The brain
One workspace, two kinds of input, one retrieval surface.
Each workspace has one brain. Everything in it is tenant-isolated, encrypted at rest with a per-workspace key, and indexed for hybrid search.
What goes in
Ingested history. Connectors pull from GitHub, Jira, Linear, Slack and Granola — backfilling years of history, then staying current through webhooks. These records are read-only over the API: tickets, prs, repos, commits, comments, discussions, people.
Agent and API context. What agents, people and your own systems write:
| Resource | What it holds |
|---|---|
documents | Free-form text you push from systems Synqed doesn't connect to. |
memories | Decisions (a choice and why) and knowledge (durable facts about a system). |
context | What's in flight right now — short-lived, expires with workspace retention. |
rules | The append-only rulebook agents load before working. |
What comes out
- Search (
POST /v1/search) runs a semantic and a keyword query in parallel and fuses the rankings, so it finds both "how do we handle retries" and an exact error string. - Lists and gets return raw records, newest first, with cursor pagination.
- Context (
GET /v1/context) returns the warm-start bundle for a new agent session.
Every read is filtered by the key's data access before anything leaves the server.
Records
Every record comes back as a JSON object with id, object (its type), source, created_at and updated_at. Connector records carry the fields their source provides — for example, a ticket from Jira includes its status, assignee and priority. Internal bookkeeping (encryption, indexing, visibility tags) is never returned.