Episodes and recall
Anton keeps a complete archive of your work sessions, called episodes. Every session is logged to .anton/episodes/ in your workspace as one timestamped file per session (named like 20260227_143052.jsonl — date and time the session started).
What's recorded
Each episode file is an append-only log of everything that happened, turn by turn:
- your messages
- Anton's replies
- every tool call it made and what came back
- output from code it ran
Long entries are truncated for size — tool inputs and outputs are capped so the archive stays compact even after heavy sessions.
Recall — Anton searches its own history
Anton can search this archive itself. Ask things like:
you> What did we do last week with the sales data?
you> What was that API endpoint we found for bitcoin prices?
Anton uses its recall tool to search the episode files — case-insensitive matching, newest results first, with an optional days-back window when the question implies a time range. It decides when to search on its own; you don't have to issue a command.
Inspecting from chat
| Command | What it does |
|---|---|
/memory episodes | Show the current conversation as a table of turns (question and answer per turn) |
/memory episodes delete <n> | Delete turn number n from the archive — the live conversation history is rebuilt without it |
Turning it off, wiping it
Episodic memory is on by default. To turn it off:
- run
/setup, choose Memory, and toggle episodic memory, or - set
ANTON_EPISODIC_MEMORY=falsein your environment or workspace.env.
When off, sessions aren't logged and the recall tool isn't offered — existing files are left in place.
To wipe the archive entirely, run /memory reset episodic (asks for confirmation).
Privacy
Episodes are written to your local disk inside the workspace's .anton/episodes/ folder and stay there. They are not uploaded anywhere. If a session touched something sensitive, you can delete individual turns with /memory episodes delete <n>, remove the session's file by hand, or reset the whole archive.