GitHub Copilot's /chronicle: A Local Memory of Every Coding Session

If you've been using GitHub Copilot CLI for any length of time, you've hit this problem: you remember solving something with a great prompt two weeks ago, but you can't find the session it happened in. Scrolling back through session lists doesn't help when the title is something generic like "refactor work." The v1.0.47 release on May 13 added /chronicle — a local query history that fixes this.

The feature is experimental. It writes every chat interaction to a local SQLite database in your Copilot config directory, then lets you query it with /chronicle followed by search terms. The database stays on your machine; GitHub does not see it.

Turning Chronicle On

Chronicle is opt-in. Run /chronicle enable inside any Copilot CLI session and the agent starts logging from that point forward. Past sessions are not retroactively imported — only sessions started after you flip it on get indexed. If you want full coverage, enable it once and leave it on.

# Turn it on
/chronicle enable

# Check status
/chronicle status

# Turn it off (logging stops, existing index stays)
/chronicle disable

Searching Your History

The search command takes free-text and matches against prompts, agent responses, and file paths touched. There are three modifiers worth knowing: --since for a date filter, --tool to filter by tool calls used, and --repo to limit search to sessions where a specific repo was the working directory.

# Find any session where you asked about JWT validation
/chronicle search "jwt validation"

# Find sessions from the last week that touched auth files
/chronicle search "auth" --since 7d

# Find sessions in a specific repo where you used /fork
/chronicle search "" --repo api-server --tool fork

# Open a result by ID to resume or read
/chronicle open ses_8f3a2b

The Workflow Tips Feature

Chronicle does more than search. After 50+ logged sessions, /chronicle tips surfaces patterns it has noticed in your work: prompts you reuse, tool calls that often follow each other, commands you start typing then cancel. The tips are local-only — they're computed from your database, not sent anywhere.

On my own setup, the most useful tip after two weeks was "You start 80% of refactor sessions with /context show. Consider making that a saved alias." Another flagged that I run the same lint-then-test pattern often enough to warrant a saved workflow file.

What Gets Logged and What Doesn't

Chronicle logs prompts, responses, tool calls, and file paths. It does not log file contents themselves — if you ran a session that read 2,000 lines of code, the database stores that the file was read, not the bytes. That keeps the index small and avoids dragging secrets from old sessions into a future search result.

Worth knowing: if your prompts themselves contain secrets (API keys, passwords typed into a session), those will be in the database. The /chronicle redact command lets you nuke specific sessions or scrub all sessions matching a pattern. Use it after any session where you typed sensitive material.

# Delete a single session from the index
/chronicle redact ses_8f3a2b

# Delete every session matching a pattern
/chronicle redact --match "prod_token"

# Nuke the whole database and start fresh
/chronicle reset

When Chronicle Pays Off Most

Three patterns where Chronicle has been a clear win: finding the right prompt template from a past session when you start a similar task, recovering context after switching repos for a week and coming back to one you haven't touched, and debugging your own workflow by looking at where sessions ran long. For the last one, --since 30d combined with sorting by turn count surfaces the sessions that ate the most time — usually the ones with the worst prompts.

If you use Copilot CLI for any work that spans more than a single sitting, turn Chronicle on. It's a few seconds of setup and pays back the first time you go looking for an old session and find it in two seconds instead of twenty minutes.