Integrations & synchronized memory
- · Each SaaS connector is OAuth-scoped—the agent sees only approved tokens.
- · Auto-fetch wakes about every twenty minutes to pull deltas into SQLite + Markdown summaries.
- · Cross-source prompts work best after Gmail (or whichever anchor source) completes one cycle.
1 · Connector model
README marketing cites 118+ integrations exposed as typed tools. Practically that means Gmail, Slack, GitHub, Notion, Stripe, Linear, Jira and others authorize via OAuth—not via your universal password.
2 · OAuth vs desktop sign-in
Hint text in the upstream docs explicitly states signing in does not grant ongoing access to every integration—you must approve each OAuth screen and can revoke scopes from vendor consoles if something feels off.
3 · Auto-fetch cadence
Once a connector is active, the core scheduler walks it on a ~20 minute cadence. That first tick is what turns raw mail or messages into Memory Tree chunks your prompts can query.
If you ask for a briefing seconds after linking Gmail, expect partial or empty answers. Wait for the first sync or force a manual refresh inside the product if the UI exposes one in your build.
4 · Token compression
README references TokenJuice—HTML → Markdown cleanup, URL shortening, ASCII normalization—so tool outputs stay billable. You do not configure it; just know summarization runs before model calls.
5 · Next steps
Read how information lands on disk in Memory tree & vault, then jump to First prompts.