Concepts
Temporal memory
The timeline, timestamps, and how agents reason about "when"
HyperMemory remembers when as well as what, in two ways: timestamps on everything in the graph, and a separate timeline — a chronological diary of what you and your agents did.
The timeline
Every meaningful change to your memory — storing, updating, forgetting, uploading — is recorded on the timeline automatically, with the tool that made it and the node it touched. Agents add their own entries for decisions, milestones and context with hm_timeline_write.
The timeline answers questions the graph can’t:
- What was I working on Tuesday afternoon?
- When did we decide this, and what happened around it?
- What happened in the last hour, before I lost context?
The timeline is never loaded automatically. Agents call hm_timeline only when they need temporal context, so everyday recall stays fast and focused.
Searching it
| Filter | Example |
|---|---|
| A period ending now | period: "7d" — 1h, 3h, 6h, 12h, 24h, 7d, 14d, 30d, 90d, 1y |
| A date range | start: "2026-10-14T13:00:00", end: "2026-10-14T18:00:00" |
| A node | node_key: "decision_jwt_auth" |
| Text | query: "deploy" |
The Timeline: every change to your memory, newest first.
In the dashboard, Timeline shows the same diary with search and period filters, refreshing live. Export it as monthly Markdown files from Account → Settings → Export.
Timestamps in the graph
Every node and edge records when it was created and last updated, and hm_get_overview reports when the graph was last written.
Freshness
Facts age. A good agent checks how old a memory is before relying on it, and corrects it with hm_update — rewriting the description to say what’s true now and keeping the history in data.