Introducing ClawMem: Your AI Agent Finally Remembers

Hazel
Hazel

I talk to my AI agent every day. We discuss architecture decisions, debug tricky problems together, go back and forth on API designs. Then the session ends, and it’s all gone.

The next morning I type “good morning” and get a blank stare from a brand new brain. Every. Single. Day.

If you use AI coding agents, you know the drill. Context is king, but context is also temporary. So I built something to fix that.

What is ClawMem?

ClawMem is a memory plugin for OpenClaw that gives your agent persistent, structured, inspectable memory. Not a giant text file. Not an opaque vector database. Actual readable records stored in a Git-compatible repository that you own and control.

Here’s what happens once you install it:

  1. You talk to your agent normally. Nothing changes about your workflow.
  2. ClawMem mirrors the conversation into a session record in real time.
  3. At session end, an AI extraction pass reads the transcript and pulls out durable facts, decisions, preferences, and lessons worth keeping.
  4. Next session, relevant memories are recalled and injected into context before the agent even says hello.

Your agent starts every conversation already knowing what it should know about you.

Show me a memory

Here’s a real one:

[42] kind:convention topic:typescript
Prefer explicit return types on all exported functions.

That’s it. A labeled record with a kind, a topic, and the fact itself. It’s stored as a Git issue on your memory repository, which means you can browse it, search it, edit it, or delete it at any time.

No black boxes. No mystery embeddings. If the AI gets something wrong, you open the record and fix it.

Why not just use MEMORY.md?

The common workaround is a markdown file that gets loaded at the start of each session. It works for a while, then:

ClawMem automates the whole thing. Memories are extracted, deduplicated (via SHA-256 hashing), labeled with kinds and topics, and recalled by relevance — not dumped in bulk.

The part I’m most excited about: team memory

Multiple agents can share the same memory repository. When one team member’s agent learns something — the client wants the proposal to focus on sustainability, the deploy freeze starts Thursday — every other agent on the team can recall that in their next session.

No Slack archaeology. No “did someone tell the AI about the deadline change?” Just shared, inspectable, structured memory that flows between humans and agents naturally.

Set it up with a few tool calls:

collaboration_org_create(login="my-team", confirmed=true)
collaboration_team_create(org="my-team", name="engineering", confirmed=true)
collaboration_team_repo_set(org="my-team", teamSlug="engineering", permission="write", confirmed=true)

Now every agent on the engineering team reads from and writes to the same memory space.

What makes it different

Inspectable by design. Every memory is a readable Git record. You can git clone your entire memory repo. Browse it in the web console. Search by kind, topic, or full text. There’s no hidden state.

Structured, not freeform. Memories carry schema labels — kind:convention, kind:lesson, kind:decision, kind:task, kind:core-fact — plus free-form topic tags for retrieval precision. This isn’t a note dump. It’s an organized knowledge base that grows with you.

Automatic with manual override. Extraction happens without you thinking about it. But you can also store, update, or forget memories mid-session whenever you want:

Remember that we decided to use Postgres for the database, not MySQL.

The agent calls memory_store and confirms it’s saved.

Zero lock-in. Your data lives in a Git-compatible repo under your account on git.staging.clawmem.ai. Clone it, export it, delete it. It’s yours.

How to get started

Paste this into any OpenClaw session:

Read https://staging.clawmem.ai/SKILL.md and follow the instructions to install ClawMem

Your agent will install the plugin, provision a memory repository, and start recording. The whole thing takes about two minutes. No separate account signup required.

After a few sessions, ask your agent:

What do you remember about my preferences?

And watch it actually answer.

What’s next

ClawMem is live and usable today, but we’re just getting started:

We’d love your feedback. Try it out, break things, tell us what’s missing.

Your agent is ready to remember. Give it the chance.