Platform
Platform Overview Maps Memory Shadow Mode Widget Governance Actions
Developers
API Docs Catalog Guardrails Dry Run Idempotency Credentials Audit Trail
Company
About Blog Contact Book a Demo
Cerebral OS Runtime

Your agent.
Our execution layer.

Already building in LangChain, AutoGen, CrewAI, or your own framework? Plug into the Cerebral Runtime API. Governance, credentials, idempotency, dry run, and a full audit trail fire automatically on every action your agent calls — across 2,800+ integrations.

View API Docs Get API Access →
2,800+
Actions in the catalog
<200ms
Average execution latency
3 endpoints
Catalog, run, and evaluate
0-config
Governance fires automatically
How it works

Three endpoints.
Everything fires automatically.

The Runtime API is three endpoints. Your agent calls actions from the catalog. Governance, credentials, idempotency, and audit trail are not features you configure — they are the execution path.

01
GET /v1/runtime/actions/catalog
Returns the full action catalog as model-agnostic JSON Schema. Add ?format=openai for OpenAI function-calling blocks. Your agent reads the catalog and knows exactly what it can call — provider, verb, inputs, outputs, and risk level.
02
POST /v1/runtime/actions/run
Your agent calls an action. Governance checks the verb against your policy. Credentials are resolved from the vault. The idempotency fence prevents duplicate execution. The action runs. Everything is logged.
03
POST /v1/runtime/governance/evaluate
Pre-flight governance check. Call before execution if you want to know whether an action will be approved, blocked, or require human escalation — before your agent commits to anything.
04
metadata: { dryRun: true }
Pass a single flag and every write is intercepted. Reads pass through. Your agent sees exactly what would happen without touching production. Same execution path as live — not a simulation.
runtime execution
# Your agent calls an action
POST /v1/runtime/actions/run
X-API-Key: sk_live_...
{
  "verb": "shopify:create_order",
  "args": { ... },
  "dryRun": false,
  "execution_id": "agent-abc-1"
}
# What fires automatically:
✓ governance check
✓ credential resolution
✓ idempotency fence
✓ audit log written
# Response
{ "success": true,
  "execution_id": "exec_9xk2p",
  "governance_status": "approved" }
Runtime capabilities

Six things your agent gets
without building any of them.

Governance, credential management, idempotency, dry run, and audit logging are infrastructure problems. Every team building production agents solves them eventually — usually after something goes wrong. We solved them once. You just plug in.

Catalog
2,800+ actions your agent can call. Model-agnostic JSON Schema by default. Add ?format=openai for OpenAI function-calling blocks.
Learn more
Guardrails
Verb-level risk classification enforced on every execution. LOW executes. MEDIUM notifies. HIGH blocks until you configure an approval gate. Policy is not optional.
Learn more
Dry Run
Pass metadata: { dryRun: true } and every write is intercepted. Reads pass through. Your agent sees exactly what would happen before anything touches production.
Learn more
Idempotency
Pass an execution_id and duplicate calls are fenced at the constraint level. The third-party API is never called twice. Retries are always safe.
Learn more
Credentials
Your agent never handles secrets directly. Credentials are stored encrypted, resolved at execution time, and injected into the action call. Your code never sees them.
Learn more
Audit Trail
Every action your agent calls is logged with verb, inputs, outputs, duration, governance decision, and execution ID. Immutable. Always on. Not configurable off.
Learn more
Framework compatibility

Works with whatever
you're already building.

The Runtime API is framework-agnostic. If your agent can make an HTTP call, it can use the Cerebral Runtime. The catalog returns model-agnostic JSON Schema by default — add ?format=openai to get OpenAI function-calling blocks ready to drop into any tools array.

View API Docs → Get API Access
LangChain
AutoGen
CrewAI
LlamaIndex
Semantic Kernel
Haystack
Custom framework
Any HTTP client
OpenAI tools format
Catalog returns tool_definition blocks in OpenAI format — compatible with any framework that accepts tool lists.
Not building your own agent?

We'll build your
AI employees for you.

Tell us what your team does manually. We map the process, build the Cerebral, connect your integrations, and deploy it to production.

Explore the Platform

Your agent.
Production ready.

Three endpoints. Governance, credentials, idempotency, and audit trail on every call. Get API access and have your agent running in production today.

View API Docs Get API Access →