API reference

The control plane is a REST API at https://api.getkaizen.io. The SDKs wrap it, but you can call it directly.

Authentication

Send a bearer token in the Authorization header.

• Agent traffic uses an API key that starts with kz_live_, created on the API keys page.
• Console and management calls use the signed-in user's token.

Authorization: Bearer kz_live_...

The action and verdict shape

Kaizen decides locally, in the client or the in-tenant sidecar, and you report the verdict along with the action it judged. The two shapes:

// action
{
  "kind": "tool_call",          // tool_call | connect | http | file | ...
  "tool": "issue_refund",
  "target": "api.stripe.com:443",
  "publisher": "stripe",
  "metadata": { "source": "egress" }   // how it was observed
}

// verdict
{
  "decision": "allow",          // allow | block
  "reason": "ok",
  "evidence": [],
  "confidence": null            // set when the reasoning check weighed in
}

The POST /v1/verdicts request body combines them, keyed by agent:

{
  "agent": "research-bot",
  "action":  { "kind": "tool_call", "tool": "issue_refund" },
  "verdict": { "decision": "block", "reason": "undeclared tool", "evidence": [] }
}

Endpoints

Report and read verdicts

Method	Path	Purpose
POST	/v1/verdicts	Report a locally-decided verdict for an action
GET	/v1/verdicts	List verdicts. Query: agent, status, decision, source, q, since, limit, offset
POST	/v1/verdicts/{id}/status	Triage: { "status": "acknowledged \| resolved \| false_positive" }
POST	/v1/verdicts/{id}/learn	Fold the action into the agent's baseline and declaration

Agents

Method	Path	Purpose
GET	/v1/agents/profiles	Every agent: activity, observation depth, declaration
POST	/v1/agents/{agent}/manifest	Declare expected behaviour: { "tools": [], "destinations": [] }
POST	/v1/agents/{agent}/settings	{ "paused": false, "judge_enabled": true }
POST	/v1/agents/{agent}/reset	Forget the learned baseline

Policy, model, and metrics

Method	Path	Purpose
GET / POST	/v1/org-policy	Org-wide blocklist / allowlist
GET / POST	/v1/llm-config	Reasoning model (BYO key), budget, confidence threshold
GET	/v1/judge/usage	Reasoning-check usage today and this week
GET	/v1/metrics/timeseries	Daily actions, anomalies, reviews

Keys, exports, and team

Method	Path	Purpose
GET / POST	/v1/keys	List and create API keys
GET / POST / DELETE	/v1/webhooks	Export destinations (webhook, Slack, Datadog, Splunk, PagerDuty)
GET	/v1/members, /v1/audit	Team and audit log
POST	/v1/members/invite, /v1/members/role	Invite and set roles (owner/admin only)

Errors

Standard HTTP status codes. 401 for a missing or invalid token, 403 when a viewer attempts a change, 404 for an unknown id, 400 for a malformed body.