Skip to main content
Connectors let SenseLab ingest events from external systems and turn them into memory entries that agents can query and learn from. An incident that paged someone at 3am becomes something the next agent working on that service can read.
Looking for the SenseLab connector inside Claude or ChatGPT? That is the other direction — memory going out to an assistant — and it is covered in Claude and ChatGPT. This page is about events coming in.
PagerDuty is the one vendor with a ready-made connector, registered under the name pagerduty and usable without writing any code. For everything else what SenseLab gives you is the plumbing: an endpoint to post to, signature verification, deduplication, and a place to hook your own transform. Wiring up a particular tool is something you do, and the routes below are how.

Three ways in


Events API

The most direct route. Any system that can make an HTTP request can write memory:
Events are stored as shared entries with agent_id="external/{source}", visible to all your agents and included in Cortex digest compilation. Good for monitoring alerts, CI/CD pipelines reporting deploy status, cron jobs pushing metrics, and anything else with an outgoing webhook.

Webhook receiver

Each connector gets an endpoint:
Requests must carry your API key as X-AMFS-API-Key. Two optional headers control routing and deduplication: If no transform matches the event type, the receiver still writes the raw payload, so nothing is silently dropped.

Signature verification

Set an HMAC secret per connector as an environment variable on the server, named after the connector:
The signature is read from X-Webhook-Signature unless the connector overrides the header name. With no secret set, signature checking is skipped.
Deduplication is held in memory with a five-minute window, so it is per-process rather than durable. Across replicas, or after a restart, a replayed event can be written twice. Because writes are copy-on-write, a duplicate creates a new version rather than corrupting anything.

API key scopes

Keys of type ingestion may write to /api/v1/events, /api/v1/entries, and /api/v1/record-context — but not to /api/v1/webhooks/*. Use an agent key for webhook endpoints.

Seeing what is registered

Dashboard → Connectors lists the connectors registered on your server and shows each webhook URL. Secrets are configured through environment variables on the server, not through the dashboard.

Writing your own connector

A connector implements ConnectorABC:
Register it as an entry point so SenseLab discovers it:
IngestionResult.action accepts "write", "context", or "skip". Today only "write" results are persisted by the webhook handler — "context" results are accepted but not yet recorded, so build around "write".

Verifying events arrived

Trigger a test event, then read the entity back:
Or open Dashboard → Memory Explorer and watch entries appear.