API
Beyond Studio, agents can be triggered and followed over HTTP. This page covers what matters to whoever integrates.
Two ways to authenticate
| Form | For what | How |
|---|---|---|
| Imaginne session | Scripts acting as a person | The same Imaginne session, in the authorization header |
| Address credential | Systems triggering one specific agent | The credential generated when creating the HTTP address |
There is no separate Agents signup or login. Whoever signs in to Imaginne is in.
The address credential is the right option for integration: it applies to one agent, has its own limits, and gives no access to Studio.
Triggering through an HTTP address
Call the address you created, with the credential and a body in the format the agent declares.
The response is immediate and is not the result:
{
"execution_id": "exec_…",
"status": "queued",
"status_url": "…"
}
The flow never runs inside the request. An automation waiting for a human approval would not fit inside an HTTP response time.
Validated input
Triggering through an address validates the input against what the agent declares. A missing required field, the wrong type, or a value outside the options are refused with no run created.
Idempotency
Send an idempotency key to make it safe to retry the request after a network timeout:
- same key, same body → returns the original run;
- same key, different body → refused.
The key does not get around rate limits.
Possible responses
| Situation | What it means |
|---|---|
| Accepted | The run was created |
| Key repeated | The original run is returned |
| Input refused | The data does not match what the agent declares |
| Unauthorized | A wrong credential or a nonexistent address — the same response for both |
| No longer exists | The address expired or was revoked |
| Too large | The body exceeded the limit |
| Limit reached | The address's rate or concurrency |
The identical response for a wrong credential and a nonexistent address is deliberate: there is no way to discover valid identifiers by trial and error.
Following a run
With the run's identifier you can:
| Action | Use |
|---|---|
| Query the state | Find out whether it completed, failed, or is waiting |
| Read the timeline | The events, in order, with cursor-based resumption |
| Cancel | Cooperative cancellation, checked between steps |
| Re-run | Creates a new run from the same version |
The timeline records semantic state. It exposes neither the model's internal reasoning nor token counts.
Triggering as a person
With an Imaginne session, a script can trigger an agent manually, giving the environment and the input. The same rate limit as a manual run from Studio applies.
Response format
Success and error have distinct, stable envelopes. Errors carry a stable code — the text may evolve, the code does not — and, where applicable, the list of fields with problems.
Isolation
Every resource belongs to an organization, derived from the credential. A resource belonging to another organization answers as nonexistent, not as "no permission".
Getting the contract
The full contract, for your installation's version, is provided by the team that administers the platform.
Next steps
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.