REST API & Webhooks
Build on top of AgentWorks
An OpenAI-compatible chat endpoint, your own agents as a completion, and search over your knowledge base — behind one scoped key. If your code already talks to OpenAI, changing the base URL is most of the work.
- €5 welcome credit
- EU AI Act & GDPR by design
- No credit card
01 — Capabilities
Everything the product does, over HTTP.
One scoped key. Call models, run your own agents, search your knowledge base, and start runs from any external event — without leaving your own stack.
Drop-in compatible
The OpenAI shape you already code against
- OpenAI-shaped chat completions, with stream: true for SSE deltas
- Choose the model per request, or list what the workspace may use
- Usage and cost land on the key’s balance, itemised per call
POST /v1/chat/completions takes and returns the request and response shape your SDK already speaks, streaming included. Point your base URL here and you get model choice, EU hosting, approval gates and per-call cost accounting without rewriting your integration.
Run agents
Run your own agents over HTTP
- Run a workspace agent as a chat completion
- Its skills, rules and knowledge bases come along automatically
- Search a knowledge base directly when you only need retrieval
POST to the agent endpoint and one of your configured agents — with its instructions, tools and knowledge attached — answers as a completion. The agent your team built in the UI is the same one your code calls.
Webhook triggers
Trigger runs from anything
- HMAC-verified inbound webhook endpoint
- Start an agent or workflow run on any external event
- Managed and RBAC-gated from your workspace
Kick off an agent or workflow run from any external system with an HMAC-verified inbound webhook — so a form submit, a CRM update, or a CI step can start work in AgentWorks.
Integrations overviewCredentials
Keys that can only do what you granted
- Six named scopes: chat, agents and knowledge, each read/write or run
- Fail-closed by design — an unscoped key is denied, never full-access
- Rotate or revoke a key without downtime; every call is audited
- Rate limits apply per plan, on top of the scopes
A key is scoped to your workspace and carries an explicit list of permissions — read, write and run are granted separately for chat, agents and knowledge. A key with no scopes is refused rather than quietly given everything.
Governance
Automated does not mean unaccounted for
- Every call recorded: which key, what ran, what came back
- Spend counts against the same balance and the same caps
- Tenant identity comes from the key, never from the request body
API traffic is not a bypass lane. Calls made with a key land in the same audit trail as anything a person did in the UI, bill to the same balance, and stay inside the same monthly caps your admins set.
02 — Reference
Key endpoints
X-API-Key: aw_ak_<your-key> # or: Authorization: Bearer aw_ak_<your-key>| Method | Endpoint | Purpose |
|---|---|---|
| POST | /v1/chat/completions | OpenAI-compatible chat completion — set stream: true for SSE |
| POST | /v1/agent/chat/completions | Run one of your agents as a completion |
| POST | /v1/knowledge/search | Search your knowledge base (RAG) |
| GET | /v1/knowledge/folders | List the knowledge folders the key may read |
| GET | /v1/models | List the models this workspace may use |
Scopes
You grant permissions explicitly when you create a key. A key without scopes is refused — there is no implicit full access.
chat:readRead chats, messages and historychat:writeCreate chats and send messagesagents:readList and read agent configurationsagents:runTrigger agent runsknowledge:readSearch and read knowledge basesknowledge:writeUpload, modify or delete knowledge
Security & compliance
Built for regulated teams
All API calls are authenticated with workspace-scoped keys, logged to the immutable audit trail, and rate-limited per plan. OpenAPI specs are shared with active customers and qualified evaluations.
- Workspace-scoped keys
- Immutable audit trail
- Per-plan rate limits
- EU AI Act & GDPR
03 — In the wild
What teams build with the API.
One scoped key — here is what teams ship on top of it.
Embed an AI agent into your own product UI without building or paying for the model layer
Let a website form, a CRM update or a CI step start an agent run over an HMAC-verified webhook
Add retrieval over your own documents to an existing app with one knowledge-search call
Swap an OpenAI base URL for an EU-hosted one and keep the code you already wrote
Start free — claim your €5 credit, no credit card
Start building on AgentWorks
Spin up a workspace, generate a scoped key, and make your first agent run in minutes.
Prefer to talk first? Request API access