Skip to content

Quickstart · 30 seconds to first decision

A few env vars. That’s it.

Point your agent’s LLM client at the Intended gateway. Every chat completion and tool call flows through Intended. Your code stays exactly the same.

02 · Point at the gateway

bash
# Mint a tenant key (mrt_…) from the console, then:
export INTENDED_KEY="mrt_live_..."
export INTENDED_TENANT_ID="<your-tenant-id>"

# Point your runtime at the per-provider gateway path:
export OPENAI_BASE_URL="https://gateway.intended.so/v1/openai"
export OPENAI_API_KEY="sk-..."

# Identify your tenant by sending the X-Intended-Key header
# alongside the base URL (via your runtime's header config).

That’s the whole change. Your existing agent code now flows through Intended. The first decision will appear in your audit chain within milliseconds.

03 · What happens next

01 · Intent compiled

LIM interprets the call

The tool call or chat completion is interpreted into a typed IntentRequest with confidence score and rationale.

02 · Authority decided

Policy + conformance evaluated

Your policy set fires. Conformance is scored. Verdict — ALLOW, REQUIRE_APPROVAL, DENY — is returned in tens of milliseconds.

03 · Token issued

Ed25519 single-use Authority Token

On ALLOW, a cryptographic token is minted, scoped, and TTL-bounded. Carried to the connector for verification.

04 · Decision sealed

Hash-linked into the audit chain

Every decision is content-addressed, sealed, and exportable. Cryptographically signed and replayable; per-tenant public-key (JWKS) verification is live.

Quickstart — Intended in 30 seconds | Intended