The problem: every AI coding session starts cold
AI coding workflows are great at fast execution, but every new session starts with partial memory. The next agent does not automatically know what tradeoffs were made, what was deferred, or which files were risky.
That context gap causes repeated analysis, avoidable bugs, and handoff friction between humans and agents.
Why chat history is not enough
- Chat history is tool-specific and often fragmented across sessions.
- It is hard to query by project concept, decision, or follow-up status.
- It may not live in the repository where future contributors actually start.
- It usually captures conversation, not a structured operational log.
Sidecar complements chat by keeping memory inside the repo in a consistent structure agents can read.
What Sidecar records
Decisions
Why an approach was chosen and what alternatives were rejected.
Work logs
What changed in a session and which files were touched.
Tasks
Follow-up work that should survive beyond a single chat.
Notes
Useful observations that do not belong in code comments.
Summaries
A refreshed project snapshot generated from structured records.
Preferences
Project-specific conventions for humans and agents.
How an agent uses Sidecar
- Read context before work with
sidecar context --format markdown. - Implement the task in code.
- Record outcomes with
sidecar worklog record. - Record new tradeoff decisions with
sidecar decision record. - Add follow-ups with
sidecar task add. - Refresh the snapshot with
sidecar summary refresh.
Example workflow
sidecar context --format markdown
# ... implement feature / fix bug ...
sidecar worklog record --done "Implemented retry logic and test coverage" --files src/retry.ts,src/retry.test.ts --by agent
sidecar decision record --title "Use capped exponential backoff" --summary "Safer under burst traffic and keeps latency acceptable" --by agent
sidecar task add "Add request timeout metrics to dashboard" --priority medium --by agent
sidecar summary refreshCTA
If you already use AI coding tools, Sidecar adds the missing local memory layer between sessions. Start with the Getting Started guide, then add a short Sidecar workflow to your AGENTS.md instructions.
FAQ
Is Sidecar a replacement for Git?
No. Git tracks what changed in source code. Sidecar tracks project memory like decisions, work logs, tasks, and summaries.
Can different AI coding tools use the same Sidecar data?
Yes. Because Sidecar lives in the repository, Claude Code, Codex, Cursor, and humans can read the same context from the same project.
Does Sidecar require cloud sync?
No. Sidecar is local-first and stores project memory inside the repo.