Skip to content

guides

Intended Documentation

Developer Integration

The developer track — submit intents, verify Authority Tokens locally, enforce them at your own service boundary, govern agent tool calls through the gateway, and handle every error the runtime returns.

3 min read

Developer Integration#

This is the build track. Every page here is grounded in the shipping code — the @intended-inc/sdk and @intended/verify packages — so the signatures, fields, and error codes you read are the ones the runtime actually uses.

There are three ways to put Intended in the path of an action, in increasing order of how much you control the integration:

  • Submit intents directly. Your service calls POST /intent, reads the APPROVED / ESCALATED / DENIED decision, and acts on the result. This is the most direct path and the foundation everything else builds on.

Intended decides and proves; your agent acts. Intended never holds your third-party keys or calls your systems.

  • Verify the token at your service boundary. The service your agent calls checks the Authority Token with @intended/verify — signature, tenant, target, expiry — and consumes its single-use nonce before acting. Make the check fail-closed.
  • Govern an agent runtime through the gateway. Point an agent's model base URL at the LLM gateway and every tool call the model emits is evaluated before it reaches your tools. No code change inside the agent.

Start here#

Make your first authorized call

API Quickstart — obtain an intended_live_ key, submit an intent, and handle the three decision outcomes with their HTTP statuses.

Verify the token before you act

Verify Decision Tokens — verify the Ed25519 Authority Token locally with @intended/verify, pin the kid, and reject on the documented failure reasons.

Enforce before you act

SDK reference — drive intents, verification and outcome reporting from your application with guard() and authorize().

All developer pages#

  • API Quickstart — submit your first intent and read the decision.
  • Agent Intent API — submit an agent action for a decision; the response is a decision, a token and a receipt; inputQuality feedback and outcome reporting.
  • Verify Decision Tokens — local Ed25519 verification, and on-robot verifiers for physical AI.
  • Enforcement SDK — the @intended-inc/sdk and @intended/verify method reference.
  • LLM Gateway — govern every agent tool call by changing one env var.
  • MCP Server and MCP Gateway — the five MCP tools, key permissions, fail-closed behaviour and where MCP data goes.
  • Runtime Integrations — choose direct / service-boundary verification / gateway for your runtime.
  • NVIDIA OpenShell & NemoClaw — compile intents into NVIDIA OpenShell policy YAML.
  • Error Patterns — every error code the runtime returns and how to handle it. </content> </invoke>
Developer Integration | Intended