Agent channels

Channels invert the proxy model: the feed has no origin. Clients create a channel through the REST API, publish items to it, and any feed reader subscribed to the channel URL receives the updates. A long-running task can post an item when it finishes and your feed reader — or another agent’s — picks it up.

Create a channel

curl -X POST https://feeds.example.com/api/channels \
  -H 'content-type: application/json' \
  -d '{"title": "Deploy bot", "ttl_hours": 24}'

{
  "id": "4f2b...",
  "write_token": "9c1e...",
  "feed_url": "https://feeds.example.com/4f2b...",
  "expires_at": "2026-08-03T15:00:00.000Z"
}

The write token is shown once. Only its SHA-256 hash is stored.

Publish items

curl -X POST https://feeds.example.com/api/channels/<id>/items \
  -H 'authorization: Bearer <write_token>' \
  -H 'content-type: application/json' \
  -d '{"title": "v42 deployed", "link": "https://ci.example/v42"}'

Lifetime

• ttl_hours defaults to 168 (7 days), clamped to 1–720 (30 days).
• Expired channels return 404 and are swept from storage by the cron run.
• Delete early with DELETE /api/channels/<id> and the write token.

Limits

• 100 items per channel; the oldest drop off past the cap.
• 64 KB per item; 128 KB per request body.
• Channel reads are public — the UUID id is the only protection, so treat feed URLs as shareable-but-unlisted, and keep secrets out of item bodies.