Your First Memory

Push a file

memax push -f ./ARCHITECTURE.md

The server:

1. Stores the content (marked processing at first).
2. Classifies it — picks a kind (semantic, episodic, procedural, or rationale) and a stability level.
3. Generates a title + summary.
4. Chunks the content semantically.
5. Embeds each chunk for similarity search.

Within a few seconds the memory transitions to active and is retrievable.

Other ways to push

Inline

memax push "We chose Postgres over Mongo — we need strong consistency"

From stdin

cat meeting-notes.md | memax push --stdin --title "Team standup 2026-04-21"

With metadata

memax push -f ./runbook.md \
  --tags "ops,incidents,runbook" \
  --hint "on-call runbook for production incidents"

There is no --category flag — the server picks kind/stability automatically (override with --title if the auto-derived title is wrong; classification overrides exist via the REST API only).

URLs (auto-detected)

memax push https://example.com/pricing

Memax stores the URL itself with content_type=link. It does not fetch or summarize the page; for full-page capture use the web app's capture flow.

Import a directory

See memax import.

Recall it

memax recall "how does the architecture work?"

Recall performs semantic retrieval — it finds content by meaning, not keyword matching. Pass --format json for machine-readable output.

Browse your memories

memax list               # recent memories, paginated
memax show mem_a1b2c3d4  # one memory in full