What it actually does
The daily friction of agent work is re-establishing context. You explained the architecture yesterday; today the agent has no idea.
claude-mem closes that loop mechanically. It captures everything the agent does during a session, compresses it with AI rather than storing raw transcripts, and injects the relevant parts into future sessions. The compression step is what makes it viable — raw session history would blow your context window within days.
It uses ChromaDB and SQLite underneath with embeddings for retrieval, and works across Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, and OpenCode.
Compared with a team memory hub, this is the individual-developer version: lighter to run, focused on session continuity rather than shared organisational knowledge. Apache 2.0, 90,000 stars, documented in over thirty languages.
Who it's for
- 01
Individual developers tired of re-explaining the same project context
- 02
Anyone working across several projects where context switching is expensive
- 03
Teams wanting session continuity without running a memory platform
- 04
People who use multiple agent clients and want memory to follow them
Where it earns its keep
- Resuming a project after two weeks without re-briefing the agent
- Keeping context across several concurrent projects
- Preserving decisions and their reasoning between sessions
- Carrying memory across different agent clients
- Avoiding repeated explanation of the same architecture
Use it, or skip it
Reach for it when
- You lose meaningful time re-establishing context each session
- Your work spans weeks and picking it back up is painful
- You switch between agent clients and want continuity
- You want compressed memory rather than a growing transcript archive
Skip it when
- You need governed team memory rather than individual continuity
- Your work is short, self-contained tasks with no continuity requirement
- You are uncomfortable with session content being stored and re-injected
- You do not want to run a local vector store
10 automations
Ideas, not tutorials. Each one is work a team does by hand today.
- 01Engineering
Project context restore
Resume a dormant project and have the agent briefed on prior decisions automatically.
- 02Engineering
Cross-client continuity
Move between Claude Code and Codex on the same project without losing accumulated context.
- 03Engineering
Decision archaeology
Ask why an approach was chosen months ago and get the original reasoning rather than a guess.
- 04Operations
Multi-project switching
Keep separate compressed memories per project so switching does not contaminate context.
- 05Operations
Handoff briefing
Export a project's compressed memory as a written brief when handing work to someone else.
- 06Finance
Session cost reduction
Measure how much context re-establishment cost before and after, in both tokens and minutes.
- 07Founders
Weekly work summary
Generate an accurate summary of what you actually did from captured session history.
- 08Operations
Memory hygiene review
Periodically review stored memories and correct anything that has since become wrong.
- 09Operations
Onboarding from history
Give a new contributor the compressed project memory as a faster route in than reading commits.
- 10Engineering
Context window budgeting
Tune how much memory gets injected so continuity does not crowd out working context.
Want one of these running by Friday?
LimeDock builds these as real workflows inside your stack — deployed to your cloud, wired into your Slack and CRM, with the code in your repo. You pay a build fee and your own API keys, nothing else.
Source
Repository stats were read from the GitHub API and reflect the last time we refreshed this entry. The editorial breakdown above is LimeDock’s own analysis — we are not affiliated with thedotmack.
https://github.com/thedotmack/claude-mem