Ask. Decide when needed. Check what happened.

Start with a request. When work needs a person or an outside service, follow the operation and read the evidence before calling it done.

Quickstart →Get a key and make your first safe request.Build the path →Follow an operation from request to receipt and story.Use the API reference →Find the route and response your code needs.

View API Start building Interactive demo

Link requests to your users. Pass user_id — a developer-defined identifier for the end user making the request (like your internal customer ID or Auth0 sub). Cadreen uses it to scope memory, conversations, and history to that person. Omit it to build a shared workspace where all requests are tenant-level. Each workspace is isolated by default.

Choose your first call

Not sure where to start? Pick the row that matches what you are trying to do. Every path starts with /api/v1/cadreen.

Choose your first call
I want to	Use	Why
Ask a question — get an answer grounded in your data	`/chat/completions`	OpenAI-compatible. Full intelligence layer — rules, memory, decision logs, tool calling. Drop into any tool that speaks OpenAI protocol.
Ask it to do something — refund, transfer, approve — with your rules enforced	`/intent`	Cadreen-native. Typed outcomes: `answered`, `needs_input`, `blocked_by_policy`, `blocked_by_missing_connection`, `executing`.
Set everything up — one call, describe what you need	`/setup`	Describe your purpose. Cadreen proposes connections, memory, policies, and webhooks. No manual endpoint choreography.
See what it can handle before you go live	`/assess`	Capability preflight. Returns what Cadreen can and cannot do for your purpose. No action taken.
See exactly why it decided what it did	`/intelligence/{trace_id}`	Full breakdown: which rules fired, what knowledge was consulted, why the gate decided what it did. Send this trace to support.

Choose your path

Two doors. Pick the one that matches what you're building. Everything else is a second step.

Swap in CadreenFastest

You already use OpenAI. Change two fields — base URL and model — keep your code. Get governance, memory, and traces immediately.

See how

Build with CadreenFull

Start fresh with the full layer — rules, memory, healing, and traces. One call to set up, and a successful answer explains itself.

Start building

Add governed tool calling

Let Cadreen call your APIs. Every call checked against your rules first.

Add memory

Teach it facts and preferences. It remembers across requests.

Add human approval

Set rules that ask you before doing something risky.

Connect via MCP

Use inside Claude Desktop, Cursor, or any MCP client.

Route to any model

BYOK, budgets, multi-provider routing. With or without Cadreen.

Operate in dashboard

See what happened, what it learned, what needs you.

For most web projects you can ignore this. It's for robots, sensors, and headsets.

Your first call — Request

POST /api/v1/cadreen/intent
Authorization: Bearer sk_cadreen_...
{"messages":[{"role":"user","content":"Refund $700 to customer 123 for order 456"}],"user_id":"cust_12345"}

Response (403)

{"id":"a2c9...","type":"blocked","status":"blocked_by_policy","next_action":{"type":"resolve_escalation","label":"Resolve the policy block","reason":"Governance policy prevented the action"},"meta":{"governance":{"decision":"handoff","reason":"Refunds over $500 need a manager's approval"}},"operation_id":null,"intelligence":{...}}

Try it live

Loading endpoints from the API reference