API reference · Messages

Send a direct message

POST /v1/conversations/{id}/messages

Free on supported networks except X, where sending a direct message costs 6 credits.

Request

Authenticate with a bearer API key. Agents calling over MCP authorise with OAuth instead and never handle the key.

curl -X POST https://api.postlake.dev/v1/conversations/{id}/messages \
  -H "Authorization: Bearer $POSTLAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"account":"…","text":"…","humanAgent":true}'

Parameters

FieldTypeRequiredWhat it is
idstringYesIn the URL.

Body

FieldTypeRequiredWhat it is
accountstringYes
textstringYes
humanAgentbooleanNoAssert that a PERSON wrote this reply. Meta allows a reply within 24 hours of someone's last message; the Human Agent tag extends that to 7 days and is granted only for replies a human composed. Never set it for an automated reply: the account carries the penalty, not the caller.

Response

Sent (200).

Example

"…"

Limits

Write operations are limited per account, and rate-limit responses include Retry-After. Do not automatically retry a write unless this operation explicitly documents idempotency.

Calling this from an agent

Agents do not call REST. They call tools. Over the hosted MCP server this operation is send_message, and the agent authorises with OAuth so your API key is never pasted into a chat or seen by the model. The reply comes back in the same normalised shape you see above, which is what lets an agent reason about it without a per-network branch.

You control what any one agent may do with it: which brands it may act for, which networks it may reach, how much it may do in a day, and whether it may spend credits you bought up front. See the MCP server.

Errors

Every failure uses one shape: a stable type, a granular code, the field at fault in param, and a plain-language fix. An agent can act on it without a person reading the message. See Errors and retries.

Check the result before retrying a write. Only operations that explicitly support idempotency can safely replay an Idempotency-Key.

Common questions

Can an AI agent do this without the REST API?

Yes. Over MCP the same job is the send_message tool. The agent authorises with OAuth, so no API key is pasted into a chat or seen by the model, and the answer comes back in the same normalised shape.

Is it safe to retry this call?

Do not automatically repeat a write after a timeout unless that endpoint explicitly documents idempotency. First read the resulting state; a second request could repeat the action.

What happens when it fails?

You get one error shape: a stable type, a granular code you can branch on, the offending field in param, and a fix written so an agent can correct itself rather than stop. Full list on Errors and retries.

Related