Skip to main content
The Leme API lets your systems do what you already do in the dashboard: put an agent to work. You send an instruction, the agent runs with the same permissions, approvals, and tools it always has, and you collect the result — by polling or by webhook. There are two ways to activate an agent from the outside:

Runs API

Full programmatic control: create a run per request, choose the instruction, follow the status, cancel, and receive completion webhooks.

Webhook trigger

A dedicated URL for external tools. The instruction is fixed on the trigger; every accepted event becomes a run.
Use the Runs API when your code decides what the agent should do on each call. Use a webhook trigger when the sender is an external tool (a form, Stripe, Zapier) and the instruction never changes.

​
Authentication

Every API request is authenticated with an API token. Create one in the Leme dashboard under Settings → API tokens.
Tokens are scoped to a project and have a mode: read_only tokens can check statuses and list runs; creating or canceling runs requires a read_write token. The token is shown once at creation — store it in your secret manager.
Treat the token like a password. Anyone holding it can act on your project within the token’s mode. Revoke it in the dashboard if it leaks.

​
Runs are asynchronous

Agent work takes seconds to minutes, so the API never blocks waiting for it. Creating a run returns 202 Accepted immediately with a run resource you can follow:
Follow the run by polling GET /api/v1/runs/:id, or register a callbackUrl and let Leme notify you — see Completion webhooks.

​
Idempotency

Network calls fail and get retried. To make retries safe, send an Idempotency-Key header with every POST:
Repeating the same key with the same body returns the original response instead of creating a second run. The same key with a different body returns 409 idempotency_conflict. Keys are remembered for 24 hours.

​
Errors

Errors always use the same envelope, with a stable machine-readable code:
Handle errors by code, not by message — messages may change, codes will not.

​
Rate limits

Requests over a limit return 429 with a retry-after header telling you how long to wait.

​
OpenAPI

The full machine-readable specification is available at GET https://app.leme.ai/api/openapi — use it to generate clients or import the API into your tooling.