Quickstart
Syok AI gives you one OpenAI-compatible endpoint in front of every AI provider you've connected. When one runs out of free credits or hits a rate limit, the next one answers automatically. Your code doesn't change.
client = OpenAI(
base_url="https://YOUR-SYOK-URL/v1", # ← the one line you change
api_key=os.environ["SYOK_API_KEY"],
)
The rest of your code stays exactly as it is. Get your Base URL and key in the app: Developer API.
Finding your way: Work and Studio
Since 10 November 2026 the app looks like Claude: a left pane with everything you can open, and the page itself on the right. Drag the thin line between them to make the left pane wider or narrower; double-click it to hide or show the pane. On a phone the left pane opens from the β° button. Every time you sign in you land on Work βΊ Get started.
- Work (the first door) is your keys and chat: Get started, New chat, Providers, Playground (Chat and Compare), API Keys, Secrets Manager, Traces (every request: who answered, which model, how long, whether Syok switched), Projects and your Recents. All chats under Recents lists every saved chat with a search box.
- Studio (the second door) is everything you make and build. The switch at the top picks Room or Code.
- Room: Create (Apps, Artifacts) Β· Collaborate (Team: Messenger and Coding space) Β· Build (Agents, Workflows, Triggers, Scheduled, Batches, Audio) Β· Context (Connectors, Files, Prompts, Skills) Β· Monitor (Evaluations).
- Code: Syok Code: New session, Files, Repos, Checks, Terminal, Virtual team, Jobs, Bring a repo, Health and Install CLI, with your recent code jobs underneath.
- Your name at the bottom of the left pane opens Settings, Usage, Get help, Install the app, Referrals & affiliates and Log out (and Upgrade plan while you are on the free plan or trial).
- The arrows next to the door name go back and forward. The search box at the top (Ctrl K or β K) finds any page or tool by name.
| Before | Now |
|---|---|
| Providers, Free AI keys guide, Tencent Cloud | Work β Providers |
| Playground, Compare AI | Work β Playground (Chat and Compare tabs). Compare is also the Second opinion agent template |
| Developer API | Work β API Keys |
| Key book | Work β Secrets Manager |
| Saved chats, chat history | Work β Recents and All chats; group them under Work β Projects |
| Syok Code, Cowork β Code | Studio β Code |
| Team chat, Rooms | Studio β Team β Messenger (chats are now called groups: New group) |
| Team coding room | Studio β Team β Coding space (Share room is now Share with group) |
| Workspace, Web Builder, Motion Studio, Game Developer, Business Suite, Creative Studio, 3D Studio, Face swap, Image to video, Film Director, the file tools | Studio β Apps (each has its own card) |
| Outputs | Studio β Artifacts (files you saved, Creative Studio pictures in this browser, and shipped code changes) |
| New: Syok Design | Studio β Apps β Syok Design: one prompt box for slides, documents, mock-ups and more; the result appears with Preview, Download and Save to Artifacts |
| Agents | Studio β Agents (Mine, Library, Software Factory) |
| Syok Lab | Studio β Workflows (List, Agent Hub, Guardrails) and Studio β Evaluations |
| Scheduled tasks, automations | Studio β Scheduled (yours and templates) Β· Studio β Triggers (run on a schedule or when an app changes) |
| Storage, storage jobs | Studio β Files Β· heavy jobs in Studio β Batches |
| Skills & plugins, connectors | Studio β Skills and Studio β Connectors |
| New: saved prompts | Studio β Prompts: save a prompt you use often; Use starts a chat with it filled in |
| Voice & translate, Live voice | Studio β Audio |
| Memory, reply length, reply language | Settings β Your Syok Β· reply language is in Settings β Account |
| Plan & billing, Refer a friend | Settings β Billing Β· Settings β Referrals & affiliates |
| Usage today | Settings β Usage |
| Signed-in devices | Settings β Account β Where you're logged in: sign out one device or all others (they must sign in again within the hour) |
| Theme | Settings β Appearance: Light, Dark or System (follows your device), text size and chat width |
| Install CLI | Studio β Code β Install CLI |
| Install the app | Your name β Install the app |
Older sections further down this page may still use the old names; the table maps them. Something went wrong with the new layout? Add ?shell=0 to the address to use the previous screen for a while, and tell us.
How Syok AI works
- You bring your own keys. For example Gemini, Groq, Mistral, OpenRouter or Cloudflare Workers AI, which all have free tiers.
- Syok AI puts them in order, either your order or cheapest first.
- It sends each request to the first provider that can take it.
- If a provider fails because of a rate limit, no credit, or being down, it automatically retries on the next one.
You pay each provider directly on your own key. Syok AI only charges for the switching service.
1. Connect providers
- Open the Syok AI app and sign in with Cloudflare. A free Cloudflare account works.
- On Providers, click Connect next to a provider. The guide opens that provider's key page for you.
- Paste the key, click Test, then Save.
- Connect at least 2 or 3 providers so there's always a backup. A good free set is Gemini + Groq + Mistral + Cloudflare Workers AI.
2. Get a developer key
- In the app, open the menu and choose Developer API.
- Switch the Developer API on.
- Click Create key. Copy it straight away: it starts with
syk_and is shown only once. - Copy your Base URL from the same page. It looks like
https://β¦workers.dev/v1.
In the examples below, replace https://YOUR-SYOK-URL/v1 with your Base URL, and SYOK_API_KEY with your key.
Call the API directly
Send a POST to /chat/completions in the standard OpenAI format. Use the model syok-auto and Syok AI picks the provider.
import os, requests
r = requests.post(
"https://YOUR-SYOK-URL/v1/chat/completions",
headers={"Authorization": f"Bearer {os.environ['SYOK_API_KEY']}"},
json={
"model": "syok-auto",
"messages": [{"role": "user", "content": "What is the meaning of life?"}],
},
)
print(r.json()["choices"][0]["message"]["content"])
print("Answered by:", r.headers.get("X-Syok-Provider"))const res = await fetch("https://YOUR-SYOK-URL/v1/chat/completions", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.SYOK_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
model: "syok-auto",
messages: [{ role: "user", content: "What is the meaning of life?" }],
}),
});
const data = await res.json();
console.log(data.choices[0].message.content);
console.log("Answered by:", res.headers.get("X-Syok-Provider"));curl https://YOUR-SYOK-URL/v1/chat/completions \
-H "Authorization: Bearer $SYOK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "syok-auto",
"messages": [{"role": "user", "content": "What is the meaning of life?"}]
}'Supported request fields:
messages, includingsystemmessages;temperature;max_tokensormax_completion_tokens;stream;toolsandtool_choice;- an image in the last user message, using the OpenAI
image_urlformat.
Using the OpenAI SDK
Syok AI speaks the OpenAI API, so the official OpenAI SDKs work as a drop-in replacement. Only the base_url and the key change.
# pip install openai
import os
from openai import OpenAI
client = OpenAI(base_url="https://YOUR-SYOK-URL/v1", api_key=os.environ["SYOK_API_KEY"])
completion = client.chat.completions.create(
model="syok-auto",
messages=[{"role": "user", "content": "What is the meaning of life?"}],
)
print(completion.choices[0].message.content)// npm install openai
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://YOUR-SYOK-URL/v1", apiKey: process.env.SYOK_API_KEY });
const completion = await client.chat.completions.create({
model: "syok-auto",
messages: [{ role: "user", content: "What is the meaning of life?" }],
});
console.log(completion.choices[0].message.content);Streaming
Set "stream": true to get a standard Server-Sent Events stream, for tools that require one. Syok AI waits until one provider has finished the whole answer before it starts sending. That way a provider switch halfway through can never mix two answers together.
stream = client.chat.completions.create(
model="syok-auto",
messages=[{"role": "user", "content": "Write a haiku about the ocean"}],
stream=True,
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")const stream = await client.chat.completions.create({
model: "syok-auto",
messages: [{ role: "user", content: "Write a haiku about the ocean" }],
stream: true,
});
for await (const chunk of stream) process.stdout.write(chunk.choices[0]?.delta?.content ?? "");Tool calling
Tool (function) calling uses the standard OpenAI tools format:
- Your code describes the tools.
- The AI replies with
tool_calls. - Your code runs the tools and sends back the results.
Syok AI skips providers that can't do tool calling and fails over between the ones that can.
tools = [{
"type": "function",
"function": {
"name": "get_weather",
"description": "Get the current weather in a city",
"parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]},
},
}]
r = client.chat.completions.create(
model="syok-auto",
messages=[{"role": "user", "content": "What's the weather in London?"}],
tools=tools,
)
print(r.choices[0].message.tool_calls)const r = await client.chat.completions.create({
model: "syok-auto",
messages: [{ role: "user", content: "What's the weather in London?" }],
tools: [{
type: "function",
function: {
name: "get_weather",
description: "Get the current weather in a city",
parameters: { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
},
}],
});
console.log(r.choices[0].message.tool_calls);Keeping one AI per chat
By default each request goes to the best available provider. To keep the same AI for a whole conversation, so the tone and style stay consistent, send a conversation ID. Syok AI then tries the provider that answered last time first, and only switches if that one fails.
client.chat.completions.create(
model="syok-auto",
messages=history,
extra_headers={"X-Syok-Conversation": "chat-123"},
)curl https://YOUR-SYOK-URL/v1/chat/completions \
-H "Authorization: Bearer $SYOK_API_KEY" \
-H "X-Syok-Conversation: chat-123" \
-H "Content-Type: application/json" \
-d '{"model":"syok-auto","messages":[{"role":"user","content":"Hi again"}]}'You can also pass it as metadata.conversation_id, or use the OpenAI user field. Long conversations are trimmed from the oldest messages so they fit each AI. Your system message and the latest message are always kept.
Auto-skills
Syok AI can add a matching skill (a SKILL.md with expert instructions) to each request, whichever AI answers. It picks up to 2 skills per message: first by keyword (for example "postgres" β the Postgres skill), then by meaning, using an embedding on your own free Workers AI. Built-in skills: frontend-design, doc-writing, internal-comms, theme-factory, mcp-builder, write-a-skill, postgres-best-practices, mysql, code-review, debugging, email-writing, summarize, translation, social-media-posts, seo-writing, data-analysis, resume-writing, image-prompts. Skills you upload in the app (Menu β AI Library β Skills & plugins) are picked the same way.
Control it per request with the X-Syok-Skills header: auto (the default), off, or up to two skill IDs, e.g. X-Syok-Skills: mysql,doc-writing. The reply header X-Syok-Skills says which skills were used (none if no skill was used). Skills are not added to tool-calling requests.
curl -i https://YOUR-SYOK-URL/v1/chat/completions \
-H "Authorization: Bearer $SYOK_API_KEY" \
-H "X-Syok-Skills: off" \
-H "Content-Type: application/json" \
-d '{"model":"syok-auto","messages":[{"role":"user","content":"Why is my MySQL query slow?"}]}'Response headers
| Header | Meaning |
|---|---|
X-Syok-Provider | Which provider answered, e.g. Groq |
X-Syok-Model | The exact model that answered |
X-Syok-Fallbacks | How many providers were skipped before this one answered (0 = the first one worked) |
X-Syok-Skills | Which skills were added to the request, or none |
X-Syok-Request-Id | A request ID (req_β¦). Include it when you contact support. |
Models
GET /models lists the models your key can use. Today that's syok-auto, which means "route across my connected providers". Your provider order and the cheapest-first setting in the app decide which provider is tried first.
curl https://YOUR-SYOK-URL/v1/models -H "Authorization: Bearer $SYOK_API_KEY"Images & video
Your developer key can also make pictures and videos, with the same automatic switching as the app. Syok AI first tries the image or video providers you connected (for example Stability AI, Leonardo or Luma), then the built-in Cloudflare image model, and then the free Pollinations image service.
import os
from openai import OpenAI
client = OpenAI(base_url="https://YOUR-SYOK-URL/v1", api_key=os.environ["SYOK_API_KEY"])
img = client.images.generate(model="syok-auto", prompt="A cozy cafe at sunset, watercolor")
print(img.data[0].url) # or ask for response_format="b64_json"# Image
curl https://YOUR-SYOK-URL/v1/images/generations \
-H "Authorization: Bearer $SYOK_API_KEY" -H "Content-Type: application/json" \
-d '{"prompt": "A cozy cafe at sunset, watercolor"}'
# Video (needs a connected video provider, e.g. Luma)
curl https://YOUR-SYOK-URL/v1/videos/generations \
-H "Authorization: Bearer $SYOK_API_KEY" -H "Content-Type: application/json" \
-d '{"prompt": "A paper boat drifting down a rainy street"}'Replies look like {"created": β¦, "data": [{"url": "β¦"}]}, and the X-Syok-Provider header shows which service made it. You can add an optional image field (a picture URL or data URL) to use as a reference. Videos can take a minute or two.
Frameworks & apps
Anything that lets you set an "OpenAI-compatible" base URL works with Syok AI. Use model syok-auto.
# pip install langchain-openai
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="syok-auto", base_url="https://YOUR-SYOK-URL/v1", api_key=os.environ["SYOK_API_KEY"])
print(llm.invoke("Hello!").content)# pip install litellm
from litellm import completion
r = completion(
model="openai/syok-auto",
api_base="https://YOUR-SYOK-URL/v1",
api_key=os.environ["SYOK_API_KEY"],
messages=[{"role": "user", "content": "Hello!"}],
)// npm install ai @ai-sdk/openai-compatible
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { generateText } from "ai";
const syok = createOpenAICompatible({ name: "syok", baseURL: "https://YOUR-SYOK-URL/v1", apiKey: process.env.SYOK_API_KEY });
const { text } = await generateText({ model: syok("syok-auto"), prompt: "Hello!" });Settings β Models β OpenAI API key:
API key: syk_β¦ (your Syok AI key)
Override base URL: https://YOUR-SYOK-URL/v1
Add model: syok-auto# pip install crewai
from crewai import LLM
llm = LLM(model="openai/syok-auto", base_url="https://YOUR-SYOK-URL/v1", api_key=os.environ["SYOK_API_KEY"])
# then pass it to your agent: Agent(role=..., goal=..., backstory=..., llm=llm)1. Add an "OpenAI Chat Model" node and connect it to your AI Agent or chain.
2. Credential to connect with β Create new β OpenAI:
API Key: syk_β¦ (your Syok AI key)
Base URL: https://YOUR-SYOK-URL/v1
Organization ID: leave empty
Save. n8n checks the credential by asking Syok for its model list.
3. Model: type syok-auto (if it is not in the list, switch the field to "By ID").
4. Turn OFF "Use Responses API" so n8n uses the standard chat endpoint.1. Add a "ChatOpenAI" node (Chat Models).
2. Connect Credential β create an OpenAI credential with your Syok key (syk_β¦).
3. Model Name: syok-auto
4. Additional Parameters β BasePath: https://YOUR-SYOK-URL/v1
(Field names can differ slightly between Flowise versions. You are looking for the
setting that replaces the OpenAI web address.)// In your OpenClaw config (models.providers):
{
models: {
mode: "merge",
providers: {
syok: {
baseUrl: "https://YOUR-SYOK-URL/v1",
apiKey: "SYOK_API_KEY", // the name of an environment variable that holds your syk_ key
api: "openai-completions",
models: [
{ id: "syok-auto", name: "Syok AI (auto)", reasoning: false, input: ["text"],
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
contextWindow: 128000, maxTokens: 8192 },
],
},
},
},
}
// Then choose it as the model: "syok/syok-auto"Other frameworks work the same way, including:
- OpenAI Agents SDK
- LlamaIndex
- Pydantic AI
- LangGraph
- Coding agents and agent harnesses β Cline, Aider, Continue, ruflo and similar
Point their OpenAI-compatible client at your Base URL. There is also a short guide to using Syok AI inside coding and agent tools.
Claude Code and other Anthropic-style tools
Your personal worker also answers the Anthropic message format (POST /v1/messages and /v1/messages/count_tokens). In Claude Code, Cline and similar tools, set the base URL to your worker's address (not ending in /v1) and use your Syok key (starts with syk_) as the token. Syok always picks the best available provider (syok-auto), whatever model name the tool asks for, and replies arrive once the answer is complete rather than word by word.
Presets, money limits and savings
These live in the app under Key book and work for the developer API too, because they are settings of your own personal Worker.
Quick presets
One tap decides which provider Syok tries first. Whatever you choose, a busy or empty provider still falls through to the next allowed one.
- My own order β exactly the order you arranged under Providers.
- Free only β only providers that have a free plan. If none is connected, Syok's built-in free AI answers. If you moved a key to a paid plan on the provider's website, that provider may still charge you.
- Cheapest first β starts with the cheapest provider and moves up when one is busy or out.
- Best quality β starts with the strongest models, then the rest in your order.
- Smart (by task) β picks a good fit for code, long documents, reasoning or quick chat.
Monthly money limit
Give a provider a monthly dollar limit. When Syok's estimate of that provider's spend reaches it, the provider is skipped until next month and the others carry on. The estimate comes from the length of each answer and a rough price list. It is not your real bill, so keep a safety margin and also set a limit on the provider's own website.
What you saved and the answer label
The Key book shows a monthly What you saved card: how many answers came from free models, how many were reused from saved answers, and the estimated saving compared with a typical paid assistant (priced at $3 per million tokens). Under chat answers you may also see a small label such as free model Β· saved about $0.01. API replies carry the same information in a saved field. All of these are estimates, never a bill.
Your last 30 days
A private list of which provider answered each time, how long it took and whether Syok switched. The words of your messages and replies are never kept in it, private chats are left out, and each day's list deletes itself after 31 days. You can clear it any time.
Limits & key safety
| Limit | Value |
|---|---|
| Requests per minute, per key | 30 |
| Daily cap, per key | You set it (default 500, up to 10,000). Resets at midnight UTC. |
| Keys per account | Up to 5 |
| Key lifetime | 30, 90 (default) or 365 days |
| Request size | 6 MB |
Built-in protection:
- The full key is shown once. Only a hash is stored.
- Revoke any key instantly, switch the whole Developer API off, or use Panic: revoke all keys.
- A key that hits its daily cap 3 times in 7 days is switched off automatically. That stops a leaked key from being abused for long.
- Developer keys can only chat and make images or videos. They can never read your provider keys, chats or settings.
- Keys only work while your trial or subscription is active.
Errors
Errors use the OpenAI error format: {"error": {"message", "type", "code"}}.
| Status | Code | What to do |
|---|---|---|
| 400 | invalid_request, invalid_json | Fix the request body (check messages) |
| 401 | invalid_api_key, key_revoked, key_expired | Use a valid syk_ key, or create a new one |
| 403 | api_disabled | Switch the Developer API on in the app |
| 413 | request_too_large | Send less than 6 MB |
| 429 | rate_limited, daily_limit_reached | Wait (see Retry-After), or raise your daily cap |
| 502 | upstream_error | Every connected provider failed. Connect more providers or top up one of them. |
Playground: files & skills
In the app's Playground, the π paperclip opens a small menu with two choices.
Attach a file or picture
- Pictures: ask about them, or use them as a reference for image and video.
- Documents: TXT, MD, PDF, Word
.docx, CSV, JSON and more. They're read in your browser and sent as text, so every AI can use them.
Skills & plugins
Auto-picked skills are used only when your message matches them (see Auto-skills). Turn the built-in ones on or off, or upload your own SKILL.md. Uploads are checked for suspicious instructions, and flagged skills stay off. Type /skill-name in a message to force a skill.
Always-on instructions apply to every Playground chat. To add one:
- Upload one or more
.mdor.txtfiles, such as aSKILL.mdfrom a plugin. Thenameanddescriptionfront matter is read automatically. - Or type a name and the instructions yourself.
Turn each skill on or off with its switch. Skills that are on are sent as a system message, so they work with every provider.
---
name: Spanish customer replies
description: Reply to customers in friendly, simple Spanish
---
Always reply in simple, friendly Spanish.
Keep replies under 120 words and end with a helpful next step.Search your chats
Find the exact words you typed or the AI wrote, in every Playground chat. Unlike asking an AI to "remember", this is a plain text search: it never guesses, skips or makes things up. Open it with the π button next to Playground, or press Ctrl+Shift+F (β+Shift+F on Mac). If you select a word on the page first, it's searched straight away.
| Type | Finds |
|---|---|
invoice | Messages that contain "invoice" (also "invoices", unless Whole word is ticked) |
invoice penang | Messages that contain both words |
"net 30 days" | That exact phrase |
invoice -draft | "invoice", leaving out messages that also say "draft" |
from:me / from:ai | Only your messages / only the AI's replies (or use the menu) |
in:title | Search chat titles instead of messages (or tick Titles only) |
Tick Match case to tell "Apple" from "apple". Click a result to open that chat: every match is highlighted, and the β² βΌ bar at the top steps through them. Export .md or .txt saves only the matching passages, not whole chats.
Privacy and cost: the search runs in your browser, so it's free and no AI or server reads your query. To make it instant, a text-only copy of your chats (no pictures) is kept in this browser. It updates only the chats that changed, and it's deleted when you sign out. You can also delete it any time with Delete it now in the search panel.
Work Studio
Work Studio (paid) turns your connected AIs into a team for real jobs. Open it from the app menu, pick a row, choose a job, fill in the short form and press Run. Keep the tab open until it finishes.
- Creative: opens Creative Studio for pictures, video and audio, plus thumbnails, banners, social posts, slide decks and the photo and video tools.
- Technical: SEO audit, GEO check (be found by AI search), QA and bug testing, UI/UX review, MVP planning, web scraping, API tester, Google Sheets helper, an API client written from a copied request (cURL, fetch or HAR, with secrets removed in your browser first), and a Code agent (see below).
- Business: Files to Spreadsheet, financial model, business plan, market research, legal drafts, email marketing, customer scripts, Amazon listing, book writing, transcription, podcast and virtual-assistant tasks, Wide Research (many angles researched at once, every number with its source), a transcript-to-summary tool (chapters, how-to steps and action items), and asking questions of your own documents.
- Connectors (MCP): use 8 free public tool servers or add your own (public https addresses only, up to 8). Read-only tools run automatically; tools that change data run only when you tick them. For apps that need a login, press Sign in with the app instead and approve it on the app's own page. The answer appears right on the page (nothing to open). If the tools only describe what is available, a Next step card offers the one more call that fetches the real result (for example, real trending videos) and runs it only when you press Run it now; the card shows the price when the tool lists it.
Projects, schedules and watches
Keep each client's brief, versions, voice sample and activity together in a project. Projects are saved in your own Cloudflare worker (your browser only keeps a copy), and you can export or import them as a file. Repeat automatically turns a plain-English request into a scheduled draft. Watch checks one read-only tool every 15 minutes to 24 hours and tells you only about new items (paid plans, up to 10). Multi-step flows pause for your approval, and Do something with this draft prepares one action that runs only after you approve it.
Code agent
The Code agent edits a GitHub repository for you. Describe a change; an AI reads the repository, edits the files, runs your tests, fixes what fails and opens a pull request. It runs inside GitHub Actions in your own repository, using your own Syok key, and needs a GitHub token you create yourself (kept only in the open tab). It merges by itself only if you turn auto-merge on and every safety check passes (tests ran and passed, at most 12 files and 600 lines changed, and no workflow, lock or secret files touched). You can also type code agent owner/repo your task in a chat to open the form pre-filled; nothing starts until you press the button.
Every result goes through an automatic quality check that looks at links, numbers and sources and shows what it found. The check is an aid, not a guarantee, so review results before relying on them. Legal, financial and similar outputs are drafts, not professional advice.
Creative Studio
Creative Studio (paid, part of Work Studio) is where you make pictures, videos and sound. It looks like a film studio: a top bar with Explore, Image, Video, Audio, Apps, Library, Recast, and a big prompt box at the bottom. Open it from Work Studio β Creative Studio, or from the AI Library β Creative. Everything runs on your own accounts, so Syok never adds a charge on top of what your provider charges.
What you can do
- Image: describe a picture and get 1 to 4 versions. Add up to four reference photos to change an existing picture. Pick the model (Nano Banana with your Gemini key, Seedream with your BytePlus key, FLUX with your fal.ai key, or the free built-in sources).
- Video: turn a picture into a clip (Image to video) or write a clip from words (Text to video), using your Qwen (Wan), Magic Hour, fal.ai or other video key. Clips usually take 1 to 5 minutes; you can leave the screen open.
- Audio: make music or a sound effect from a sentence (needs your fal.ai key), plus on-device tools: voiceover, recorder, trim, volume, speed, pitch, equalizer and more. The on-device tools never upload your files.
- Apps: 60 one-step apps and tools (headshot, restore an old photo, upscale, remove background, trim a video, slide deck and others). Photo, video and design tools that used to live in Work Studio β Make open on top of the studio, and the β brings you straight back.
- Relight, Speak (a photo that talks), Recast (swap a face or head in a video), and your Library (everything you made, saved on this device).
Connect your fal.ai key (step by step)
- Go to fal.ai and create an account. fal.ai is pay-as-you-go: add a few dollars of credit in Billing. A picture usually costs a fraction of a cent to a few cents and a short clip or music track a few cents, but check fal.ai's own price on each model page.
- Open fal.ai/dashboard/keys and press Add key. Choose the API scope, give it a name like "Syok" and copy the key (it looks like two long codes joined by a colon). fal.ai shows it once.
- In Syok open Creative Studio, press the π key button (Your providers), find fal.ai, press Connect, paste the key and press Save.
- Syok checks the key with fal.ai. If fal.ai refuses it you see a plain message and nothing is saved. A saved key is stored encrypted in your own Worker and only a masked version (first and last characters) is ever shown.
- Pick fal.ai in the Image or Video model list, or use Audio β Make music. Long jobs (video, music) keep running on fal.ai; Syok checks every few seconds for up to about 8 minutes.
If something fails: "didn't accept this key" means create a new key; "no credit left" means top up fal.ai Billing; "doesn't know this model" means pick another one.
Voices and speech with Fish Audio (your own key)
Creative Studio β Audio has a Voices & speech card powered by Fish Audio. You type words, pick one of thousands of voices, and hear them spoken. It can also turn a recording into text. Fish Audio has a free tier, and it runs on your own account.
- Create a free account at fish.audio, then open fish.audio/app/api-keys and create an API key. Copy it.
- In Creative Studio press π (Your providers), find Fish Audio, press Connect, paste the key and press Save. Syok checks the key and stores it encrypted; it is never shown again.
- Open Audio. Type what it should say. To change the voice, type a word like "calm" or "narrator" in Find a voice, press Search voices, press βΆ Hear to listen and Use this voice to pick one.
- Leave the model on Free model to spend nothing, or choose S2.1 Pro for the best quality (uses your Fish Audio credit). Press βΆ Speak, then β¬ Download to keep the MP3.
- To write down a recording, choose a sound or video file (up to about 7 MB) under Turn a recording into text and press Write it down.
Tip: add feelings in square brackets, like [whispering] or [laughing]. "Check my balance" shows the credit left on your Fish Audio account. If it says "no credit left", add credit at fish.audio or keep using the free model.
Copy a voice, design a voice, and add a voiceover to a video
These use the same Fish Audio key as above. Copying and designing a voice use a little of your Fish Audio credit, so they ask nothing until you press the button.
- Copy a voice: open Audio, scroll to Copy a voice from a recording. Give it a name, then upload 10 to 30 seconds of one person speaking clearly, or press π€ Record and speak. Press Copy this voice. It becomes your chosen voice. Only copy your own voice, or one whose owner agreed. The copy is private on your Fish Audio account.
- Design a voice: under Design a voice from words describe it (for example: warm, calm woman in her 30s). Press β¨ Design voices, play the samples, and press Keep as my voice on the one you like.
- Delete a voice you made: in Find a voice tick the box for your own voices, then press π Delete and confirm with Really delete?.
- Voiceover on a video: under any video you made (or in Library) press π£ Add voiceover. The Audio & Music tool opens on Add a voiceover. Type the words and press π£ Make the voice (or add your own recording), pick how loud the voice is and how quiet the original sound stays, then press βΆ Run. The result is a new MP4 made on your device.
Where did Make go?
Work Studio β Make now only lists the few tools Creative Studio does not have yet (slideshow, video watermark, banner, event flyer, quote card, share card, pin, and the Qwen tools). Everything else moved into Creative Studio, and Make will be phased out over time.
Work Studio Tools: 127 everyday file tools
Inside Work Studio (paid plan), press π§° Tools on the left rail. You get a toolbox of 127 small tools for the jobs you would otherwise do on a converter website. The tools window has a Category menu, a Tool menu and a search box (on a phone these become two tabs, Tool and Results). Each tool has an info panel and a work area, and every button explains itself when you hover over it. The β at the top takes you back to Work Studio.
Your files never leave your browser. The tools run on your own device, nothing is uploaded to Syok, and nothing costs credits. The first time you use a heavy tool your browser downloads the engine it needs once and keeps it (for example the video engine is about 31 MB, speech-to-text models 40 to 250 MB, the text-to-speech voices and the background-remover model each download on first use).
- PDF Editor (13): merge, split, organize, compress, OCR, edit, annotate, sign, fill forms, redact, protect and more.
- Business & Brand (11), Privacy & Security (4) and Writing & Text (6): everyday document, brand, privacy and text helpers.
- Online Converter (35): Word, Excel, PowerPoint, EPUB, images, archives (open ZIP, RAR and 7z; make ZIP, TAR or 7z with an optional password), units, time and more.
- Image Editor (8), Video Editor (20) and GIF / Memes / Fun (7): edit, resize, compress, AI background remover, trim, crop, merge, add text or music, make GIFs and memes. Video also has a Frame picker (scrub a video and grab frames, or auto-grab 1 to 200 evenly spaced, as PNG, JPG or a ZIP) and Image has Photo to 3D relief (STL) for 3D printing.
- Audio Editor (15): see the section below.
- Record / Speak / Subtitles (8): voice, camera and screen recorders, text to speech, speech to text and subtitles (soft track or burned in).
What works in other languages
- OCR turns scanned PDFs and photos into searchable PDFs in many scripts, including Chinese, Japanese, Korean and Arabic. The same OCR feeds scanned PDF to Word, Excel and EPUB.
- Subtitles in Chinese, Japanese, Korean, Arabic, Thai and other non-Latin scripts burn into the video correctly.
- Text to speech has about 33 voices in many languages with a language picker and speed control. Japanese and Thai use your device's own voice, so they can be played but not saved as a file. Voice quality is basic for Arabic, Hindi, Korean and Bengali.
- Word and PowerPoint to PDF keep the original look and the text stays searchable. The result is a picture of each page plus an invisible text layer, so it is not editable text.
- RAR files can be opened but not created; use 7z or ZIP to make an archive.
Tools that make a new file never change the original. Results can be downloaded or passed straight into the next tool.
Audio Editor
All 15 audio tools run on your own device: nothing is uploaded and nothing costs credits. They are in three groups.
- Edit sound: cut, volume, speed, pitch, reverse, equalizer, join, convert, compress and ringtone maker.
- Analyse & split a song: Key, tempo and chords; Stem splitter (vocals, drums, bass, other); and Vocal remover (karaoke).
- Make music: Audio to MIDI (turns a melody into a MIDI file), Mashup maker (blends two or more songs, matching tempo and key) and Loop maker.
AI or Quick engine. The stem splitter, vocal remover and mashup maker have an Engine choice. AI uses the Spleeter model running in your browser. On the test song it separated much more cleanly than Quick (measured signal-quality scores of 9.3, 6.4, 8.1 and 5.8 dB for vocals, drums, bass and other, against 3.0, -1.0, 3.2 and 1.2 dB). The first time you use AI your browser downloads about 79 MB of models from Syok's own storage and keeps them, so later runs start straight away. AI is cleaner than Quick but not studio-perfect.
Length. AI handles up to 20 minutes (8 minutes on phones and small devices). Quick handles 5 minutes for stems and 8 for the vocal remover. Every splitting tool has From / To boxes, so any longer file can be done in parts; the mashup maker has them per song, and Audio to MIDI works on up to 12 minutes of the part you select. Loop maker and Key, tempo and chords have no length cap.
Files. Stems come out as MP3 (192 kbps, about 4 MB for a 3-minute stem) or, if you prefer, WAV (about 30 MB). A Stop button cancels a long run. Not yet tested on real phones, only on emulated ones.
Results shelf
Whatever a tool produces is kept in a Results shelf inside your browser for 3 days, where you can preview, download or delete each item, or delete everything with a two-step confirmation. Nothing is uploaded. Clear your browser data and the shelf is emptied.
Compare AI
Compare AI asks the same question to up to 5 of your connected AI providers and shows the answers side by side. Use it to cross-check a solution, a piece of code or a decision: if different AIs agree you can trust the answer more, and if they differ you know what to double-check. Open it from Menu β AI Library β Compare AI.
- Tick providers. Only providers you have connected are listed. Choose 2 to 5 (the maximum is 5; the other boxes switch off until you untick one). Your choice is remembered on this device.
- Type one question and press Ask. Every ticked provider is asked at the same time, using your own key. Press Stop to cancel.
- Read the answers. Each card shows the provider, the model used and how long it took. You can Copy an answer, mark your favourite, or copy everything.
- Compare the answers (optional). One of the providers that answered writes a short plain-words summary: where they agree, where they differ, and what to double-check. This uses one more request.
Nothing is swapped. Normal chat moves to the next provider when one is rate-limited. Compare AI never does, so you never see the same AI twice. If a provider fails, its card says why (rate-limited, key rejected, payment needed, timed out) and you can retry just that one. Each ticked provider uses one request from its own limit.
Technical note: the app calls POST /v1/compare on your personal Worker once per ticked provider, with {provider, messages, temperature?}. It uses only that provider with your saved key, with no switching and no built-in AI fallback, and returns {ok, provider, model, text, ms} or {ok:false, outcome, message}. It counts toward the same per-user chat limit as normal chat.
Agents
Build your own AI helpers that use your keys. Open the app menu β π€ Agents, press οΌ Agent, and either describe the helper in a sentence (Create with AI) or start from a blank one. Each agent has three panels: Settings (model and price ceiling, skills, tools such as web search or reminders, knowledge, checks, limits, approvals), Instructions, and a Playground to try it. Press Publish to make a version live; older versions can be compared and restored.
Knowledge (your documents)
- Add knowledge: type a note, or upload
.txt,.md,.csv,.json,.htmland now PDF, Word (.docx) and Excel (.xlsx/.xls) files up to 10 MB (they are turned into text first). Add tags to keep documents grouped. - Retrieval settings: how many passages the agent reads (Top 2/3/5/8), an optional Reranker (Cloudflare's free model re-orders the best passages), and Entity links (passages that name the same people, companies or codes as the question move up).
- β Auto-tune tries several settings against the agent's Tests (the "contains" checks) and keeps the one that finds the answer most often. πΈ Entities shows the names and codes found in your documents.
Everything is stored in your own Cloudflare account, not on Syok's servers.
Syok Code: the AI engineer on your own keys
Syok Code plans, checks and ships changes to a GitHub repository using the AI keys you connected. Open it from the AI Library (the π Syok Code card). It is its own workspace now, not part of Agents, and no longer has a Menu entry. Read the overview at syok-ai.com/syok-code.
The screen: three panes
π» Code on the left, π Live preview in the middle and π¬ Ask advisor on the right. Each pane has two buttons at its top-right corner: β shrinks it to a thin strip, β€’ makes it fill the screen. Press the same button (or the strip) to bring it back. On a phone the panes are stacked.
First steps
- π Connect β GitHub: on GitHub open Settings β Developer settings β Fine-grained tokens, choose only the repositories Syok Code may change, and allow Contents and Pull requests (read & write) plus Issues, Checks and Actions (read). Paste the token yourself. It is encrypted in your own storage. (The Software Factory uses the same token.) Press π How to get your key for these steps in a pop-up.
- Pick the repository (and a branch, if not the default) at the top.
- π Files: filter and open any file. HTML, Markdown and SVG files appear in the Live preview. HTML pages get their own CSS and JavaScript files put in, and run in a sealed box. You can also type a web address there.
π¬ Ask advisor: the οΌ button
Press οΌ next to the question box to add Files, a Folder or Images to your question. They are read inside your browser and sent to the advisor only when you press Ask. Up to 12 text or code files (about 80,000 characters in total) and 4 pictures; folders skip node_modules, .git and similar. Pictures you add are combined with the screen picture into one image. Files you attach count as read, so the reference check will not hold back an answer for naming them.
π§ͺ Self-check: no gate, no ship
The π§ͺ Self-check tab asks the advisor a few fixed test questions (quote a line from an attached file, add two numbers from an attached file, count to 30 without getting stuck, follow a one-word instruction, and not invent a file it cannot see) plus three Build (council) tests, and scores the answers. The Build tests give one council member a tiny built-in project and check that it replies in the right format, only edits files it was shown, keeps the change small and does not add risky code such as eval, using the same checks that guard a real Build. Run it once and press Save as βbeforeβ, make your change to the AI side, then run it again. Syok compares the two runs and tells you Safe to ship or Hold this change: the overall score must go up by at least 5 points (or, when the βbeforeβ score was already 95% or more, must not go down), and no area (attachments, loops, format, honesty, council) may drop by more than 2 points. Free models can take a few minutes per question, so you can press Stop at any time. The βbeforeβ score is kept in this browser only. When new test questions are added, run it once more and save a fresh βbeforeβ: only questions present in both runs are compared.
π Loop guard: stuck replies are skipped
Small and free models sometimes get caught repeating the same sentence or list item until they run out of space. Syok now notices this in any chat reply, treats it like a failure, and tries your next model instead. The model that looped goes to the back of the line for 10 minutes (it is not removed). If every other model fails, you still get the answer with the repetition cut down to one copy, never an error. Normal replies, numbered lists, tables and code are not affected.
π³ Build: three models vote
- Describe the change in plain words and pick a mode: Quick (low-cost models), Careful (mixed, the default) or Max (your strongest models).
- Press Plan it with the council. Three members write the change at the same time: Council (temperature 0.1), Debate (0.2) and Red team (0.3). A separate lead model picks one and says why. A member that fails three times in a row rests for 10 minutes. Every run has a trace showing each step and how long it took.
- Check the vote, the plan, the π‘ ship gate report (25 security checks on the changed lines) and the coloured diff. Preview the plan's version shows it in the middle pane.
- Choose what happens after your tests pass (merge by itself, wait for your OK, or never), then press Approve & ship. Syok Code makes one commit on a new branch, opens a pull request and watches your CI. If the tests fail it tries a fix. A critical security finding gets one automatic fix attempt; if it is still there, nothing is committed.
π¬ Ask advisor and the file-reference check
Ask anything about the repository. Every file path and version number in the answer is compared with the files Syok Code read for that answer. If it names one it did not read, the answer is held back, the real file is fetched and the model answers again (up to twice). Anything still unconfirmed is marked β . The badges under each answer show this. With πΈ Show the advisor my screen ticked, a picture of the Syok page you are looking at is sent along (password boxes are left out). It goes to your Gemini key first, or to the free built-in vision model.
The small π Ask about this screen button at the bottom-right of every screen does the same thing anywhere in the app.
π©Ί Health, π¦ Jobs and undo
- Health: a free score from 0 to 100 (no AI is used) over 17 things: tests, CI, committed secrets, lockfile, licence, risky code and more. Fix turns a problem into a Build task.
- Jobs: every shipped change with its pull request and test state. Buttons: Merge, Check tests now, Close, and β© Undo. Undo opens a revert pull request; your history is never rewritten. To hand a GitHub issue to one of your agents instead, use the Software Factory link there.
π§βπΌ Team: a virtual engineering team and a one-click sprint
The Team tab turns Syok Code into a small engineering team of 24 specialists, adapted from gstack by Garry Tan (MIT licence). They are grouped in the order a real sprint runs: Think (Office Hours, Design Consultant), Plan (CEO, Engineering, Design and DevEx reviews, Auto Planner), Build (the council), Review (Staff Engineer, Design, DevEx, Bug Investigator, Second Opinion, Security Officer, Health), Test (QA Lead, QA Reporter, Performance), Ship (Release Engineer, Deploy Captain, Canary Watcher, Technical Writer) and Reflect (Retro, Lesson Keeper).
- Run one role: press Run on a card. Write your goal in the sprint box first if the role needs one. Results are saved per repository for 60 days.
- Run the whole sprint: one click runs ten steps in order (Build only if you tick the box). Each step is one call to your own AI keys and sees what the earlier steps said. If a step fails the rest continue.
- AI quality: Free is cheapest but small models can skim; use Smart or Best for Review and Security. A result that only describes what the model would do is flagged as no review.
- Honest limits: every role is read-only. It sees only the few files it is shown and cannot run code, open a browser or watch your live site. Every finding must quote your code, and the quote is checked against the file; unchecked ones are marked unverified and lower the confidence score. Nothing is committed or deployed by a role; the Builder only makes a draft, and shipping still goes through the ship gate and your approval. gstack parts that need a real browser or VM (browse, live QA, canary monitoring) are not included.
- From other tools: the same team is on the Syok Code MCP server as
syok_team_roles,syok_team_run,syok_team_results,syok_team_sprint_startandsyok_team_sprint_step.
π¬ Film Director: videos written as code
The 25th team member, Film Director, writes a short silent or narrated video as a few web pages (using the open-source OpenFilm format, MIT licence) and has GitHub render it into an MP4 for free. It does not use video-generating AI: the pictures are designed animation (text, shapes, charts), so it suits explainers, titles and product clips, not realistic footage.
- One-time setup: make a private GitHub repository for your films that contains the Render film workflow (the Syok films template), and give your Syok GitHub token Contents: Read and write and Actions: Read-only on it.
- In the Team tab pick that repository, describe the video, choose a quality (Best gives the most reliable result), and press Film Director.
- It saves the film on a new branch named
film-β¦. Nothing is merged and your main branch is never touched. - Press π Check the render. After about a minute you see four preview pictures from the film and a link to the run page, where the film download is the finished MP4.
- Voice-over (optional, 13 languages): ask for a voice-over in your description and name a language if you want one. The model writes
narration.txtand picks a voice invoice.txt, and a free Piper voice reads it over the film. Voices: US English (Kristin, Norman, LJ Speech), British English (Cori), German, French, Dutch, Italian, Spanish (lower quality), Swedish, Norwegian, Ukrainian, Greek and Czech. Only voices trained from scratch on public-domain, CC0 or CC BY 4.0 recordings are offered, so commercial use is allowed; the render summary tells you the licence and the credit to give (the German, French and Dutch voices need "Multilingual LibriSpeech" credited when you publish). Most other Piper voices were fine-tuned from a research-only voice, so they are left out on purpose. Not available: Chinese, Japanese, Korean, Malay, Indonesian, Hindi, Thai, Arabic, Spanish (good quality), Portuguese, Russian and Turkish; narration falls back to English or no voice.
Limits: up to 4 shots and 30 seconds, no external pictures, fonts or videos. Rendering uses your GitHub Actions minutes (about 1 minute per film). The model sometimes leaves a requested detail out, so look at the preview pictures before you share a film.
π¦ Bring a repo: use Claude Code skills from GitHub
Many open-source repositories ship Claude Code skills: text files (usually SKILL.md) that teach an AI how to do one job well, for example a Godot game-development guide. Open Syok Code > π¦ Import, paste the GitHub address and press Check this repo.
- Honest report: every item is marked Works, Works after conversion (it mentions commands or Claude Code tools), Needs a terminal (hooks, local MCP servers) or Can't be used. Scripts are counted but never run. The report also shows the repository's licence.
- Safe by default: Syok only reads text through your own GitHub token. Items containing hidden characters, "ignore previous instructions" tricks or key-stealing wording are switched off.
- Use: pick an imported skill, say what you want, and it runs through your own AI keys as advice: it cannot run commands, the game editor or tests, and gives you the exact steps to run yourself. Long skills are cut down to the sections that match your question, and the most relevant reference files are added. Tick "Let it read my selected repository" to show it your own code.
- Invented-name check: code names in the answer that are not in the skill text or your files are listed so you can verify them. Free and Balanced models can invent function names, so use Smart or Best for anything important.
- Big repos: a scan reads the first 30 items. The rest show as Not checked yet and are checked when you import them (30 at a time). To list just one part, paste a folder address such as
github.com/owner/name/tree/main/plugins/python-development. Agents (agents/*.md) and commands (commands/*.md) import and run the same way as skills (tested on wshobson/agents). - Remote MCP servers: if a repo lists a web (https) MCP server, press Add to Work Studio in the report. If it needs a sign-in or key, Syok tells you to finish in Work Studio.
- Limits: up to 10 imported repos; a scan reads the first 30 items; hooks, stdio MCP servers and scripts are not supported yet. Remote (https) MCP servers must be added by hand under Work Studio.
βΆ Checks: run your tests on GitHub and let the AI fix failures
Syok Code has no computer of its own, so Syok Code > βΆ Checks borrows GitHub's free one (GitHub Actions) in your own repository.
- What happens: Syok makes a throwaway branch
syok/check-...(never your main branch), adds a small workflow file if you have none, and GitHub runs your project's build and tests. Projects withpackage.json(npm), pytest,go.modorCargo.tomlare detected; other kinds report "nothing to run". - AI fix (optional): if you tick it, the AI reads the end of the error log and the files involved, changes the code on that branch, and GitHub runs the checks again, up to 3 rounds. It may not edit workflow files or secret files, and every changed file is scanned for secrets first.
- You stay in control: nothing is merged. When the checks pass, press Open a pull request and review the changes yourself.
- Needs: a GitHub token with Contents and Workflows (Read and write) and Actions (Read), and Actions switched on for the repository. Each round uses roughly one to two minutes of your free Actions time. If the token has no Workflows permission, Syok says so; you can also add the workflow file yourself.
- Limits: it cannot run game engines such as Godot, browsers or anything needing secrets; the AI sees only the log end and up to seven files.
β¨ Terminal: type commands instead of clicking
Syok Code > β¨ Terminal is a command bar: you type a short command and it runs the matching Syok Code feature. It is not a real shell, so programs like npm, git or python do not run in it. Real commands (your tests) run through check, which uses GitHub for you.
- Look around:
repos,repo owner/name [branch],ls [text],cat path,open path. - Ask the AI:
health,ask your question,roles,team role your goal. - Skills:
scan https://github.com/owner/namereads a repo for Claude Code skills (then finish in the π¦ Import tab),packslists what you imported. - Run and fix:
checkruns your tests on GitHub and waits for the result;check fixlets the AI repair failures (up to 3 rounds) on a temporary branch.stopstops waiting; the run continues in βΆ Checks. - Move around:
jobs,go files|build|health|team|import|check|jobs|connect,clear,help. - Keys: Enter runs, Up/Down brings back earlier commands, Tab finishes a command, file name or repository.
- Limits: one command at a time; if you type a real program name it explains instead of running it; the screen keeps the last 400 lines and clears when you reload.
π§° Sandbox: run real commands (npm, python, gitβ¦)
Syok Code itself has no computer, so for real commands it can borrow a temporary one from Daytona using your own Daytona key (optional). Daytona gives $200 of free credit; a 10-minute session costs roughly 1 to 3 cents.
- Create a key at app.daytona.io under API Keys.
- In Syok Code > π Connect > 1b Β· Sandbox, paste it and press Connect. It is encrypted in your own storage and never shown again.
- In β¨ Terminal type
sandbox start. Your repository is copied in (your GitHub token is never put inside the sandbox). - Run commands with
sh npm test, or just typenpm install,npx vitest run,python -m pytestand so on. - Type
sandbox stopwhen you are done.sandbox statusshows what is running.
- Safety: one sandbox at a time, 30 minutes at most, each command up to 90 seconds, output cut to the last 12,000 characters, and Daytona also deletes it after 10 idle minutes.
- Fast fix loop:
sandbox testruns your project tests inside the sandbox (npm, pytest, Go or Rust are detected).sandbox fixdoes the same and, when tests fail, asks the AI for a repair of up to 3 files, scans it for secrets, writes it into the sandbox and runs the tests again (2 repairs at most). A round takes seconds to about 2 minutes instead of waiting for GitHub each time. Nothing reaches your repository until you runsandbox push. Free models can be wrong, so read the log and thesandbox changeslist first. - Saving your work:
sandbox changeslists what changed;sandbox push your messagesaves changed text files to a new branch namedsyok/sandbox-β¦in your repository and gives you a link to open a pull request. Your main code is never touched. Workflow files (.github), env files, binary files, files over 200 KB and secrets are refused, at most 40 files, and file permissions (executable) are not kept. - Limits: repositories over 25 MB are refused; the sandbox has internet access, so only run code you trust.
Use it from Codex, Claude Code, Cursor or another tool
Syok Code is also an MCP server at <your personal Worker>/v1/code/mcp. Make a developer key (see above), keep it in the SYOK_API_KEY environment variable, and copy the ready-made lines from π Connect. For any other tool, press οΌ Add and use type Streamable HTTP, the address and the header Authorization: Bearer <syk_ key>. Tools: syok_repos, syok_tree, syok_read_file, syok_ask, syok_health_scan, syok_security_scan, syok_plan, syok_ship, syok_job_status, syok_revert.
Good to know. Syok Code is part of the paid plan (and the free trial). GitHub only for now. One plan changes up to 6 files. Plans are kept 2 days and traces 7 days, in your own storage.
Syok Lab: data flows, dashboards and AI governance
Syok Lab works like Dataiku, so people who know it will find the same layout: a top bar with Projects Β· Flow Β· Datasets Β· Dashboards Β· AutoML Β· Agent Hub Β· LLM Mesh Β· Govern, a Flow canvas in the middle and an Actions panel on the right. Open it from the app menu β π§ͺ Syok Lab (or AI Library β Syok Lab). New here? Press π Sample project on the Projects page: it creates customer feedback data, a cleaning step, an AI classifier, a join, a summary table and a dashboard you can explore.
Flow: datasets and recipes
- Add a dataset (οΌ Dataset): upload a CSV, TSV or JSON file (save Excel sheets as CSV first), paste a table, or link a Google Sheet published as CSV (it refreshes when the flow is built). Datasets are drawn as violet squares.
- Add a recipe: select a dataset and pick a recipe in the Actions panel. Orange circles are visual recipes: Prepare (17 steps such as rename, fill empty, split, formula, change type, extract date, remove duplicates), Filter, Join, Group (count, sum, average, median and more), Pivot, Sort, Stack, Distinct, Sample, Top N and Split. Pink circles are AI recipes: Prompt, Classify, Summarize, Extract, Translate and Sentiment, one AI call per row. Indigo is Embed, which sends rows into an agent's knowledge.
- Preview before you run: every recipe shows its result as you edit. AI recipes have Test on 3 rows so you can check the answers before paying for the whole table.
- βΆ Build all runs every recipe in order inside your browser. To rebuild part of the flow, select a recipe and press βΆ Run this + downstream. The bar under the menu shows progress, including how many AI answers have come back. One slow AI provider cannot hold up a build: an AI call that takes over 60 seconds is marked on that row and the build carries on.
- β° Schedule & budget: rebuild hourly, daily or weekly in your own Cloudflare Worker (up to 5,000 rows per scheduled run) and set a spending limit for AI recipes.
Datasets, charts and dashboards
- Datasets tab: Explore shows the table with a validity bar per column (green = valid, red = wrong type, grey = empty); click a column header for its profile. Charts makes bar, line, area, pie and scatter charts; Add to dashboard pins one. Schema lets you change a column's type.
- Dashboards tab: tiles for numbers (count, sum, average), charts and tables. β edits a tile, β removes it.
β¨ Cobuild: say it in plain English
Press β¨ Cobuild and type what you want, for example "count the comments per channel and add a bar chart of it to the dashboard". Your AI proposes the recipes and dashboard tiles; nothing changes until you press Apply. Then build the new dataset.
π€ AutoML: predict a column
Teach the computer to predict one column from the others, for example which customers will leave, a house price, or a ticket's category. It works like Dataiku's AutoML prediction. Training runs in your browser tab, so it is free, needs no AI keys and uploads nothing.
- Start: open the π€ AutoML tab and press οΌ New prediction model, or select a dataset in the Flow and press π€ AutoML (also under Actions β Machine learning).
- Design: pick the dataset and the column to predict. Syok decides whether it is a classification (one of several answers, up to 20) or a regression (a number); you can change it. The Features table shows every column with a reason: IDs, long free text, and columns that are mostly empty or always the same are switched off. Tick any back on if you disagree. Choose how to judge the models: ROC AUC, accuracy, F1 or log loss for classification; RΒ², MAE or RMSE for regression.
- Train: five kinds of model are tried: a baseline (always the most common answer or the average, the score to beat), a linear model, a decision tree, a random forest and gradient boosting. 20% of the rows are held back, and every model is scored only on those rows it never saw. Up to 10,000 rows are used; at that size, training takes a few seconds.
- Result: a leaderboard sorted by your metric, with β on the best model and a plain message saying whether it beats the baseline. Which columns matter shows how much the score drops when a column is shuffled.
- Model: open any model to see its scores, a confusion matrix (classification) or a predicted-versus-real chart (regression), and a What if⦠form: change the inputs and the prediction updates at once. ⬠Report (.md) downloads a model report.
- π Deploy to Flow adds a teal Predict recipe that adds a
predictioncolumn (andprobability, how sure it is, for classification) to any dataset, then builds it. It runs every time the Flow is built, including scheduled builds. To swap in a better model later, open it and press Update the deployed model.
Good to know. The leaderboard is saved with the project, but trained models stay in the browser tab; after a reload press βΆ Retrain (same settings give the same scores). A deployed model is saved inside its Predict recipe (up to 400 KB). Scores come from one 20% test split, not cross-validation, and there is no automatic hyper-parameter search, so treat them as a good first estimate. Tested in a desktop browser.
π§ Agent Hub
One page for all your agents: runs, success rate, cost, typical and slow response times (p50/p95), π/π, and drift (success rate of the last 7 days compared with the 7 days before, flagged when it drops). Scheduled evaluations re-run every agent's tests daily or weekly and send a push notification when the pass rate falls below your floor. Compare models runs the same tests on two presets side by side.
π‘ LLM Mesh: guardrails, cache and limits
Rules that sit in front of every model call made with your keys (API, chat and agents), kept in your own Cloudflare account:
- Personal data: e-mails, phone numbers, card numbers (checked with the Luhn rule), IBANs, IP addresses, US social security numbers and secret keys. Choose off, detect (log only), redact (the model sees
[EMAIL_1],[CARD_1]instead) or block. - Prompt injection: catches tricks like "ignore previous instructions" or "reveal your system prompt".
- Moderation with Llama Guard on Cloudflare's free allowance, topic boundaries (words that must not appear), and your own custom rules (words or patterns).
- Response cache: an identical request (exact) or the same question in other words (semantic) is answered from the cache, with no model call and no cost.
- Rate limits: requests per minute for your whole account and per Developer API key.
- Test a prompt shows what would be blocked or redacted and exactly what the model would see. A log and daily counts list every block, redaction and cache hit.
π Govern
- Registry: every agent with its live version, owner, routing, tools, risk level, tests and usage.
- Audit trail: who changed what and when (agents, guardrails, flows, keys), the last 500 events, downloadable as CSV.
- Publish policy: optionally require a minimum test pass rate, the LLM Mesh guard switched on, a filled-in model card and a sign-off before a new agent version can go live. On the agent's Publish bar, tick I have reviewed this version to sign off.
- Model cards: one document per agent (purpose, data, model, limits, evaluation, guardrails, history) you can download as a Markdown (.md) file.
Costs. Syok Lab itself is free. Visual recipes, charts and dashboards run in your browser. AI recipes and Cobuild use your own AI keys (cheapest first, as usual). Scheduled builds, guardrails and evaluations run in your own Cloudflare Worker on its free allowance.
Limits. Up to 30 projects, 40 datasets per project and about 8 MB per saved dataset (bigger tables stay in the browser only). AI recipes send at most 2,000 rows per run. Tested with real AI on the sample project in a desktop browser. Phones, Safari and very large files have not been tested yet.
What's new (1 to 6 November 2026)
- π Syok Code (new, its own workspace in the menu): three models vote on each code change, answers are checked against the files actually read, a security gate runs before every commit, plus a health score, one-click undo and an MCP server for Codex, Claude Code and Cursor. A π Ask about this screen button now sits on every screen. See Syok Code and syok-ai.com/syok-code.
- π€ AutoML (new): train five kinds of prediction model on any dataset in your browser, compare them on a leaderboard, try "what if", and deploy the best as a Predict recipe in your Flow. See AutoML.
- Fix: text columns in saved Syok Lab datasets now load back exactly as saved (before, "yes" could come back as "true" and a code like "007" as "7").
- Syok Lab (new): Dataiku-style projects with a visual Flow, 11 visual recipes, 6 AI recipes, charts, dashboards, scheduled builds and β¨ Cobuild, plus Agent Hub, LLM Mesh guardrails and Govern. See Syok Lab.
- Agents knowledge: PDF, Word and Excel uploads, tags, reranker, entity links and auto-tune. See Compare AI Agents.
- Connectors: the answer appears directly on the card; if a connector only lists its tools, a Next step card suggests one follow-up call with a Run it now button. Nothing runs until you press it.
- Agent tools:
POST /v1/responses(OpenAI Responses format used by Codex) now works alongside/v1/chat/completionsand/v1/messages(Claude Code). The agent-tools guide has new Goose, Codex and Claude Code sections. Tested on the wire; the programs themselves were not run. - Creative Studio: new look with Explore, Image, Video, Audio, Apps (one-photo templates, including a Product category), Relight, Speak, Recast and Library. Pollinations free image generation is always on; paid engines use your own keys.
- Free photo editing: product shots, packshots, restore and similar apps work with no key through a Cloudflare model (photos under 512 px, 1024x1024 result, shared daily allowance; face swaps are not offered).
- Bring your own key: fal.ai (FLUX, LTX Video, Wan, Kling, music, MiniMax H3, Seedance 2.5, GPT Image 2.5), Qwen Image (Alibaba key), Fish Audio voices (speak, clone, design, voiceover) and WaveSpeed character swap in Recast. Costs go to your own accounts. These were built and tested against fake keys; a real-key run has not been confirmed.
- Offline Brain: Spark X2.6 (4B and 1.7B) added for browsers with WebGPU; others keep the Basic brain. Free, on your device.
- Voice: opt-in Live typing on your device (Moonshine, English, about 28 MB once); with no internet, Offline Brain voice chat lets you speak and hear the answer.
- Work Studio Tools: 127 tools in 10 groups, the new tools window and Results shelf, AI stem splitting and the Audio Editor above.
Honest limits: the real-key runs above, Safari, iPhone and Firefox behaviour of the newest tools, and on-device runs on real phones have not been verified yet.
Team: Messenger and Coding space
Open Studio β Team. Messenger is the team chat (chats with several people are called groups; press New group). Coding space is a shared place for a team's code and conversation. One person is the owner: the team space's files, history, tasks and chat are stored in the owner's own Cloudflare worker. Everyone else is a member and keeps using their own AI keys and credits; Syok does not store a member's keys there.
- Owner: open Studio β Team β Coding space, press Open my team space, type a member's name and add them. An invite line is shown once; copy it and send it to that person. You can add as many members as you like and revoke any of them at any time.
- Member: open Studio β Team β Coding space, choose I am a member, paste the invite line and press Join. The invite opens only this team space, never the owner's keys or account.
- Work together: shared files with versions (if two people save at once, the second person sees a conflict instead of overwriting), history you can restore from, tasks, a live activity feed, and Ask my AI to change a file using your own provider. On a task, Run tests runs your repository's tests in GitHub Actions (set up once through a pull request) and posts pass or fail in the team space.
- From Code: in Studio β Code, Share with group opens Team with a new group for that repository.
Team chat
Message options: tap and hold any message (on a computer, right-click it or use the small β in its corner) to react, Reply, Forward to up to 5 chats, Copy, Star, Pin, Edit your own text message within 15 minutes, see Info (who has read it), Delete for me or Delete for everyone (tap twice to confirm). On a phone you can also swipe a message to the right to reply. β at the top lists your starred messages. Stars and "delete for me" are saved with your account in the room, so your phone and computer always show the same.
Ticks, like WhatsApp: π while sending, one grey β when the room has it, two grey ββ when it has reached everyone in the chat, and two blue ββ when everyone has read it. If a message cannot be sent you see β with Retry. Info shows who read it and who received it, with times.
Select several: choose Select in a message's options, then tap (or tap and hold) other messages. The bar at the top can star, forward, copy, delete for you, or delete for everyone (it asks first).
@mentions: type @ and pick a person from the list. Their name is highlighted, the chat list shows @ next to a chat where you were mentioned, and a mention still notifies someone who muted the group.
Pins last 24 hours, 7 days or 30 days (you choose when pinning); the pin bar shows the time left, and you can unpin any time.
The chat is for your team only: there are no phone numbers and it cannot message anyone outside the room. Messages arrive live, and you can reply, edit within 15 minutes, delete for everyone, react with emoji, pin up to 20 messages and search. Ticks show sent, delivered and read (they turn blue only when everyone has read). You also see who is typing and who is online. Send photos, files and voice notes of up to about 600 KB each (photos are shrunk automatically), turn on π notifications, and see an unread count on the π₯ button. Next to each person who is online, π starts a voice call and πΉ a video call.
Chat tab and group chats
On a phone, the chat is its own tab, π₯ Chat, next to AI Library and Playground. A red number at the top right of the tab shows how many messages you have not read, across all your chats, and it keeps updating while you are on another tab. On a wide screen the chat stays on the right of every screen and you can minimize it.
The chat opens on a list of your chats: the built-in Team chat (everyone in the room) plus any private groups you belong to, each with its last message, time and unread count. Anyone in the room, including the owner, can press οΌ New group and name it. Picking people is optional: you can create the group on your own and add people any time later from Group info, with the invite link or by e-mail. The person who creates a group is its admin. Open a group and tap its name to see Group info: admins can rename the group, add or remove people and make or remove other admins; everyone can mute it or leave; an admin can delete it. Replies, edits, reactions, pins, photos, files, voice notes, read ticks and search all work inside each group, and only the people in a group can see its messages or get its notifications (muted groups stay quiet).
Add people to a group: invite link, e-mail or contacts
Groups use an invite link or an e-mail address, not phone numbers. Open Group info as an admin:
- Invite link: press Create invite link, then copy or share it. Each group has its own link. Reset makes the old link stop working at once and gives a new one, and Turn off removes it. When someone opens the link they see a Join card on their chat list.
- Add by e-mail: type or paste e-mail addresses and press Add. People already in the room whose e-mail matches join straight away. Other addresses wait in a Waiting to join list (up to 50) and join automatically when that person is in the room and saves the same e-mail. E-mail them the link opens your own mail app with the invite link filled in; Syok does not send e-mails itself.
- Contacts: Phone contacts lets you pick people from your phone (Android Chrome only), and Import contacts file reads a .vcf or .csv file, for example one exported from Google Contacts. The addresses are filled into the Add box so you can check them before pressing Add.
- Your e-mail: every member can save their own e-mail in the box at the bottom of the chat list (the owner can also type it when adding a member). It is kept in the owner's worker and is never shown to other members.
An invite link or e-mail only adds people who are already members of the room. Only the owner can add a new member to the room itself.
Good to know: calls go straight between browsers and use free connection helpers, so on some strict networks a call may not connect. The chat keeps the latest 500 messages. The owner's worker stores everything in the room, so it is not end-to-end encrypted. Cloudflare's own free-plan limits for the room count against the owner's account.
3D Studio
Make, edit and print 3D models. Open it from AI Library β π§ 3D Studio, or just attach a .glb, .gltf, .obj, .stl or .ply file in the Playground. New to 3D? Press File β New project to start with a flat base plate, then add shapes.
The screen. The menu bar on top (File, Edit, Add, Tools, Generate, Export, View) holds every action. The icon column on the left switches the right-hand panel: βοΈ Edit, π¨ Look, π§° Tools, β Library, π¬ Animate, β Print check, β¨ Generate, π₯ Team, π€ Export. The buttons on the 3D view turn wireframe, grid and auto-spin on or off, reset the camera, take a screenshot and go full screen. Drag to turn the model, scroll to zoom, tap a part (in the view or in the Parts list) to select it. Sizes are always millimetres.
- Say what to change (βοΈ Edit): βmake it 150 mm tallβ, βlengthen the handle by 20 mmβ, βdrill a 6 mm hole through the middleβ, βmirror itβ. Your own AI turns the sentence into a short list of exact edit steps and the app applies them β the AI never touches the mesh, so every edit is exact and identical on every device.
- Exact size: type a width, height or depth for the selected part (or the whole model). βKeep proportionsβ scales the rest to match. Files without real units (STL, OBJ) can be fixed by typing one known measurement.
- Edit history: every change is listed. Press β next to any step to remove just that step (later steps are replayed; any that no longer make sense are dropped and you are told). Undo (Ctrl+Z), Redo (Ctrl+Y) and βback to originalβ are always there.
- Look (π¨): ready-made materials (metal, chrome, gold, silver, copper, glass, plastic, rubber, wood, matte, ceramic), a colour picker, and sliders for metal, roughness and see-through. Picture textures from glTF/GLB files are kept and are included when you export GLB, the web page or AR.
- Tools (π§°): Hole / cut (round hole, square notch or ball dent β type the spot, or press βPick spot on modelβ and tap it), Combine (join, cut one part out of another, or keep the overlap β exact and print-ready), Hollow out (set a wall thickness, optional drain hole), Split (separate loose pieces of a one-lump file, or cut a part in two at a flat plane), Sculpt (raise, dent, smooth, flatten or inflate by dragging; each drag is one undoable step; best on parts with lots of triangles) and Measure (tap two points for the exact distance).
- Library (β): rounded box, gear, nut, bolt, pipe/ring, donut, pyramid, star, stairs, 3D text, box, cylinder, ball and cone, each with size boxes. Shapes are added on the floor beside your model unless you type a position; then cut holes in them or combine them.
- Animate (π¬): make a selected part spin, slide or grow/shrink, by itself or when tapped, with repeat and back-and-forth. Press Make interactive page to download one
.htmlfile that plays the animations and reacts to taps; put it on any website. - Print check (β ): pick your printer (or a size close to it) and press Check. It flags open holes in the surface, inside-out faces, very thin walls, floating pieces and models bigger than the bed. Tap a problem to find the part. Exporting 3MF and letting your slicer repair is a good second step.
- Export (π€ or the Export menu): GLB (web, AR; metres), STL, 3MF (one object per part, with colours), OBJ and PLY (all millimetres); Blender (a GLB plus a ready Python script β in Blender open the Scripting tab, open the .py, press Run, keep both files in one folder); screenshot (PNG), a 6-second turntable video (WebM), a high-resolution studio render (PNG with soft shadows), and AR (spin it on screen, or place it on your table from a phone: Chrome on Android or Safari on iPhone). Links to Blender, Spline, Tinkercad and render farms are in the Export panel; Syok does not send your model to them.
- Share with your team (π₯): the file goes into your Team: Messenger and Coding space (stored on the owner's own Cloudflare, up to 12 MB). Teammates open it from the chat or Team models, edit it and save a new version. A version is the list of edits, so nothing is overwritten. If two people edit at once the second save is refused with a clear message.
Generate (β¨, or π· Photos β 3D):
- Exact shape, free, on your device: add photos from the front, back, sides, top and bottom (front + one side is the minimum), press βοΈ to remove backgrounds, type one real measurement, and the shape is carved from the silhouettes. Good for simple objects, boxes, bottles, brackets; it cannot see hollows or undercuts.
- Tripo AI (best quality, textured): type a short description, or leave the box empty to use your first photo. It uses your own Tripo account and its free credits (the free plan includes a monthly amount). Create a key at platform.tripo3d.ai β API Keys (it starts with
tsk_), paste it into the Tripo box and press Check key to see your credits. The key stays in your browser; your own Worker forwards the calls. It takes a minute or two, then the model opens with its texture. AI guesses shape, not size β type the real height in Exact size afterwards. Your personal Worker updates about 4 minutes after a new version. - Microsoft TRELLIS (MIT licence) on Hugging Face: a full textured guess from one or several photos, on Hugging Face's free GPU Space with your free Hugging Face token (kept in your browser only). It takes 1β3 minutes and the free queue can be busy.
- Tencent Hunyuan 3D with your own Tencent Cloud keys (see Storage and helpers).
Limits: up to 2 million points per model in the browser; live sculpting works on parts up to 60,000 points; the interactive web page needs a model under 12 MB and loads its 3D engine from the internet; holes and combine are exact on closed (watertight) parts, and the studio reports an open surface instead of guessing. Turntable video needs Chrome or Edge.
Video editing tools
Open AI Library β Work Studio β Creative. The Creative page has three groups: πΌ Image (AI Photo Editor, Image Generator, Upscaler, Watermark Remover, Background Remover), π¬ Video (the editing tools below) and π¨ Designs (thumbnails, banners, posts, slides).
The video tools run entirely on your own device: the first time, your browser downloads the editing engine (FFmpeg as WebAssembly, about 31 MB) and keeps it; after that everything works offline, nothing is uploaded and nothing costs credits. Files up to 400 MB. Every result can be downloaded, or sent straight into the next tool with πͺ Keep editing this result.
- πΈ Image to video β one photo or a set of photos becomes a video: Motion (made on your device, free) gives each picture a slow zoom or pan, crossfades between pictures (fade, dissolve, slide, wipe, through black, circle), 9:16 / 1:1 / 4:5 / 16:9 at 1080p / 720p / 480p, and optional music with fades; Animate with AI sends the first picture and your description to the video provider you connected β or, with your own free Magic Hour key, to an exact model you pick (Kling, Veo 3.1, Sora 2, Seedance, Wanβ¦) β and brings the clip back; there's also a negative prompt box ("avoidβ¦") and a β¨ prompt enhancer that rewrites your idea with your own chat AI (the result says whether the provider used the picture or only the description).
- βοΈ Trim & cut β keep a part (Instant: no re-encoding, no quality loss, starts on the nearest keyframe; Frame-exact: re-encodes) or remove a section from the middle.
- π Resize & reframe β 9:16 for Reels / TikTok / Shorts, 1:1, 4:5, 16:9, 4:3 or keep the shape; fill by cropping, blurred bars or black bars; 1080p / 720p / 480p.
- β© Speed & reverse β 0.25Γ to 4Γ (voices keep their pitch) or play backwards (clips up to 90 s).
- π Audio tools β mute (instant), extract the sound as MP3 / M4A / WAV, add background music (looped, with fades, under or instead of the original), louder / quieter, or level the loudness to β16 LUFS.
- π Convert & compress β MP4 (H.264 + AAC, plays everywhere), WebM (VP8 + Opus) or GIF (two-pass palette, up to 60 s); a target size in MB (e.g. 25 for email, 16 for WhatsApp); rotate and flip.
- π¬ Captions, text & logo β burn an .srt / .vtt into the picture, add a title or caption (position, size, colour, dark box, show fromβuntil), or stamp a PNG logo in a corner with a chosen size and opacity.
- π§½ Remove watermark β for footage you own: press Grab this frame, drag a box over the logo or text (several boxes allowed, or type the position), then Fill in (painted over from the surrounding pixels β best for small stamps), Blur, or Crop that edge off; optionally only between two times.
- πΌ Frames & thumbnails β one frame at a time you pick (PNG or JPG), a contact sheet of the whole video, or a burst of stills as a ZIP.
- β Join clips β several clips in the order you set; different sizes and formats are matched to the first clip (centred on black), sound is added where a clip has none.
- π Face swap (video) β put a new face onto a person in your clip. A cloud tool: it runs on Magic Hour with your own key (new accounts get free credits, no card) β the clip is processed by Magic Hour, not on your device.
- π Character swap (video) β the character in your clip is replaced by the person from your photo, keeping the clip's motion, expressions and camera (MoCha, a 2026 research model). A cloud tool: it runs on WaveSpeed with your own key and pay-as-you-go credit (about $0.04 per second of clip at 480p). Find it in Creative Studio β Recast.
- π§β𦱠Head swap (video) β the person from your photo replaces the person in the clip, head and all (or "animate": your photo moves like the clip). A cloud tool: it runs on Akool (character swap) with your own key; new accounts get free credits.
- ποΈ Reference video to video β your clip is the reference and Magic Hour redraws it in an art style you pick (Ghibli, Pixar, pixel art, oil paintβ¦), on your own key and credits, with an optional prompt for extras.
Limits: it is your computer or phone doing the work, so long 4K videos take a while (a progress bar and a Stop button are always there); the VP9 encoder is not used (it crashes in the browser build), so WebM uses VP8.
Storage: your files in your own Cloudflare
Open AI Library β Storage. It works like the file manager on your computer: folders, upload (files or whole folders, or drag and drop), download, copy link, rename, move, delete and search. My files is the same screen.
Your files stay in your own Cloudflare account. Syok does not keep your files. When you sign in, Syok creates a bucket called syok-files in your Cloudflare account (Cloudflare includes 10 GB of R2 storage free each month). If Storage shows the Use my own Cloudflare R2 button, press it, press Authorize in the Cloudflare window, and come back. You never paste any keys.
- Limits β 10 GB, 90 MB per file, up to 100,000 files. Pictures are shrunk to WebP on your device before upload. Share links are public and impossible to guess, and are served from your own Worker address.
- Bring your own cloud (optional) β connect Backblaze B2, Amazon S3, Cloudflare R2 (with keys) or Wasabi, one at a time. After that, videos and files of 20 MB or more are saved there. When your free 10 GB is 80% full, Syok reminds you on every app open and every save.
- Tools β Task List, Cloud Transfer, Cloud Sync, Cloud Backup and Website Backup run inside your own Worker, so your files and keys never pass through Syok. Team Transfer and Email Migration are being built; Remote Upload and Shares are coming soon.
- Good to know β files over 30 MB are skipped by the tools; the tasks copy 12 files at a time; Google Drive, OneDrive, Dropbox, MEGA and FTP are coming soon.
Everyday helpers
These are free unless noted.
- Memory and Private chat: short notes about you help any AI pick up where you left off. Manage them in Menu β Memory. A πΆ Private chat is never remembered.
- Voice: speak instead of typing, and Listen to read answers aloud.
- Translate: pick a language and see which AI translated.
- Reminders and scheduled tasks: set them in plain words. Notifications come through your browser, on a best-effort basis.
- Web search: live results with a daily allowance (20 per day on the free plan, 200 on paid). You can add your own search key.
- Places and contacts: save addresses and contacts, and ask for nearby places, weather and exchange rates.
- Report button: flag a wrong answer and Support fact-checks it.
Offline Brain
Offline Brain (paid) downloads a small AI model to your device (about 1 to 3 GB; a 1B model on phones, 3B on computers) so you can chat without internet. It is less capable than online AIs. Answers are marked π§ . Turn it on in Menu β Offline Brain.
Game Developer tools
Open AI Library > Game Developer > π§ͺ Tools. Everything runs in your browser except the pieces that say they need a key or the engine. Ideas are inspired by Fuser's game-development tools; the Godot guidance credits vl4dt/godot-skills and haxqer/godot-skill (both MIT).
- π Check: a free checker for Godot 4 mistakes (old names, yield, signals, missing colons, broken scenes and shaders). It is not a full GDScript parser. Fix the safe problems repairs what is certain. Ask the AI to fix the rest sends the remaining problems to your chat and you press Apply to my code.
- β Real Godot: runs the real Godot 4.3 engine in your Syok Code sandbox (your own Daytona key) and shows the engine's own errors for scripts, scenes and shaders, then can ask the AI to fix them and re-run. The sandbox is a paid-by-you service and stops after 30 minutes; press Stop when done.
- π Sprites: turn pictures (or an AI sheet on a plain green or colour background) into transparent frames, a sprite sheet and a Godot
SpriteFramesfile with animations. - 𦴠Rig: cut a character into 10 parts and get a 2D skeleton scene (Skeleton2D, Bone2D, walk and idle animations). It does not rig 3D models; for 3D characters use Mixamo.
- π§ 3D: picture to 3D model through the free TRELLIS Space on Hugging Face. It needs your own Hugging Face token (Keys page).
- π§© TileSet: makes a TileSet with collision, plus level code for TileMapLayer and the older TileMap.
- π¦ More: input-map and 2D and 3D starters, and six C# templates (they compile against GodotSharp 4.3).
- π― Look lock & AI maps: make several views of one character that keep the same look (each later view uses the first picture as a reference), and ask the AI for a height map to build better normal, roughness and occlusion maps. Needs your image key; if your provider refuses reference pictures the tool tells you and uses the same words instead.
Honest limits: AI pictures can still differ between views; the checker finds the common mistakes but Godot's Play button has the last word; Godot reports one script syntax error at a time; the 3D tab was not tested end to end without a Hugging Face token; there is no automatic 3D rigging.
FAQ
Is Syok AI a model provider?
No. Syok AI routes your requests to providers using your own keys, and switches automatically when one runs out.
Does it cost extra per request?
No. There's no per-token markup. New accounts get a 14-day free trial of everything, no card. After that you can stay on the free plan for as long as you like, or pay $5/month or $30/year for the paid features. Your providers charge only for what you use on your own keys, and free tiers cost nothing.
What's free and what's paid?
Free plan (always): chat with every AI provider you connect, with automatic switching when one runs out; the built-in Cloudflare AI; cheapest-first and my-own-order routing; saved chat history, search, pins and replies; Memory and Private chat; voice, translation, reminders and scheduled tasks; saved places and contacts; and a daily web-search allowance.
Paid plan adds: Work Studio (including Creative Studio and tool connectors); Offline Brain; a bigger daily web-search allowance; priority support; image and video generation; attaching pictures and documents to chats; smart routing; auto skills (18 built-in plus your own uploads); adding your own AI providers; the Developer API (your own OpenAI-compatible key, including tool calling, and the Anthropic-style endpoint for Claude Code); and the members' list of free video tools.
Referral program
Share your link from Menu β Refer a friend. You earn 30% of every payment a referred user makes for their first 12 months for each referred user. Every 100 paying referrals raises your rate 5%, up to 50%. Commissions are paid by hand on the 1st of each month via Wise or PayPal, once they're past a 14-day hold and add up to at least $20. Refunded or charged-back payments don't earn commission.
Which providers are free?
Google Gemini, Groq, Mistral, OpenRouter's free models and Cloudflare Workers AI all have genuinely free API access, with no card needed. Three more (OVHcloud AI Endpoints, the Kilo Code free gateway and LLM7.io) need no key at all: type free when asked. Syok AI supports 24 providers in total. Stack several so you rarely run out.
Where is my data?
Your keys (encrypted), chats, memory, reminders and saved places are stored in your personal worker in your own Cloudflare account. Work Studio projects are stored in your personal worker (your browser keeps a copy), and the Team coding room's files and chat are stored in the room owner's worker. Web search goes through our central service and a search provider, with a shared cache that holds no account information. See the Privacy Policy.
Need help?
Email [email protected], or use the support chat in the app. Include the X-Syok-Request-Id if you have one.