Documentation
Connect your apps to Worksi AI Control: the quickstart, API keys, the OpenAI‑compatible gateway, providers and routing, the SDKs, MCP, OpenTelemetry, policies, end‑user plans and incidents.
Get started
Connect
- 03Gateway APIThe OpenAI‑compatible endpoint and Anthropic’s Messages API: endpoints, attribution and response headers, the request pipeline, errors, providers and the simulator.
- 04Providers & routingConnect any of 24 provider presets or your own OpenAI‑compatible server, pick the provider per call with a route prefix, add fallback models and read every cost in USD.
- 05Python SDKTrace runs, steps and tools, record outcomes, request human approvals and honor operator controls from Python.
- 06TypeScript SDKWire the OpenAI and Anthropic Node.js clients to the gateway and trace runs, steps and tools with typed errors and approvals.
- 07MCPConnect Claude Desktop, Cursor, VS Code and your own agents to Worksi over the Model Context Protocol, and govern any MCP server through the MCP gateway.
- 08OpenTelemetrySend OTLP/HTTP traces from existing OpenTelemetry instrumentation. GenAI semantic conventions are mapped automatically.
Govern & operate
- 09Policies & approvalsGovern models, spend, tokens, rates, tools and data capture with policies, roll them out in monitor mode, and keep people in the loop with approvals.
- 10End‑user plans & quotasMeter and limit AI usage per end user of your app according to the plans you sell: free, pro or anything else.
- 11Incidents & alertsHow detectors open incidents, what the evidence pack and diagnosis contain, how to recover, and how alerts reach email, Slack and webhooks.
Questions
More questions? Email info@xeoo.io.
Do I have to change my code?
Two lines: point your OpenAI client’s base URL at the Worksi gateway and use a Worksi API key instead of your provider key. Code written for the Anthropic SDK changes the same two values. Your provider keys stay with Worksi, encrypted. The Worksi SDKs are optional and add workflow-level tracing, approvals and outcomes.
Which models and providers work?
Any of 24 provider presets: OpenAI, Anthropic, Gemini, Mistral, xAI, Groq, Together AI, OpenRouter and Azure OpenAI; the Chinese providers DeepSeek, Qwen, Kimi, GLM, MiniMax, ERNIE, Hunyuan, Doubao, BytePlus and StepFun; and Ollama and vLLM for models you run yourself. Any other server that speaks the OpenAI API works too. The gateway serves Chat Completions, the Responses API, embeddings and streaming, and Anthropic’s Messages API for Claude. A built‑in simulator gives deterministic, free responses for tests and demos.
Do you charge for model tokens?
No. You bring your own provider key and your provider bills you directly. Worksi never resells or marks up model tokens; our plans cover the control plane only.
What happens to my prompts and responses?
You decide per environment: store previews in full, store them with sensitive values redacted (the default), or store metadata only — hashes, token counts, cost, timings and errors, with no text at all.
How does the free trial work?
Every account starts with a 3‑day trial with Pro features and no credit card. Confirm your email to lift the trial caps. To continue after the trial, contact us or use Request upgrade in the console and we activate your plan.
Can AI assistants and agents connect over MCP?
Yes. The Worksi MCP server gives agents tools to trace runs, check policies and request approvals, and gives people tools to work incidents from Claude Desktop, Cursor or VS Code. The MCP gateway governs any MCP server you register.
Start with the quickstart.
Create a key, point your OpenAI client at the gateway and open your first trace, in about ten minutes.
# Governance → API keys (the key is shown once)export OPENAI_API_KEY="aiw_sk_live_…"export OPENAI_BASE_URL="https://gateway.worksi.ai/v1"