Claude-Mem by thedotmack

What it is

Claude-Mem is Alex Newman’s memory add-on for Claude Code and other coding agents. A new session starts blank, so you explain the project again. Claude-Mem records what the agent does, has a model turn that into short notes, and loads the recent ones when the next session opens.

1. Hooks capture your prompts and the agent’s tool calls
2. A local worker sends them to your memory provider
3. Short notes land in SQLite under ~/.claude-mem
4. The next session opens with the recent notes

Who it suits

Developers who work with a coding agent on the same codebase day after day and are tired of the recap. Long refactors and projects that run for weeks gain the most.

If your sessions are short and unrelated, the memory adds cost without much gain.

Where to use it

One installer covers most agents, and a few get a line of their own.

• Claude Code, Cursor and Codex CLI: the general installer pre-selects the ones it detects
• OpenCode: its own line, for OpenCode 1.3.4 or later
• T3 Code, Antigravity CLI, OMP and Pi: the installer’s --ide option, one host per command
• DeepSeek: a plugin of its own, through the same option
• Grok Bot: it has no host hooks, so claude-mem watches its chat log files
• OpenClaw gateways: an install script from cmem.ai, worth reading before you run it
• The local viewer: saved notes, sessions and settings at /viewer
• Claude Desktop: a skill that searches the same memory from Claude Desktop chats

How to set it up

Get it saves a zip of Mindket’s setup notes in your downloads folder. The plugin, its hooks and its worker install from thedotmack’s repository.

1. Run the installer in a terminal and keep the agents you use selected.
2. Sign in through your browser when it asks, or pass an explicit provider flag to skip that.
3. Pick a memory provider: CMEM Pro, your own OpenRouter or Gemini key, or your Anthropic plan.
4. Restart your agent so the hooks load.

npx claude-mem install

Claude Code can take the plugin from the maker’s marketplace instead, followed by a restart. OpenCode 1.3.4 or later has a line of its own.

/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem

npx claude-mem install --ide opencode

The README warns that npm install -g claude-mem installs only the SDK, with no hooks and no worker.

How to use it

After the install it runs by itself, and earlier notes appear when a session starts. To look something up, ask in plain words. The agent works through three tools, cheapest first.

The README shows the calls like this:

// Step 1: Search for index
search(query="authentication bug", type="bugfix", limit=10)
// Step 2: Review index, identify relevant IDs (e.g., #123, #456)
// Step 3: Fetch full details
get_observations(ids=[123, 456])

A few commands keep the install healthy.

Example

You open a new session and ask about the last one.

What bugs did we fix last session?

The agent searches its memory, reads a short list of matching notes with their IDs, then fetches the full notes it needs. It answers with the fixes it found, and each note carries an ID you can find again in the web viewer.

For newcomers and experts

If you are new. A coding agent forgets everything when a session ends. Claude-Mem writes short notes about what happened and hands the recent ones to the next session. Install it, work one session as usual, then open a new one and ask what you did last time.

For experts. Six lifecycle hooks, from SessionStart to SessionEnd, each run one claude-mem command through a launcher the installer puts on your PATH. A worker managed by Bun summarizes what they capture through your memory provider into SQLite, and a Chroma vector database adds semantic search beside keyword search. Retrieval reads the index before full notes, which the README credits with about ten times fewer tokens.

The costs are a summarizing call on every session and your activity going to the provider you pick.

Good to know

What the hooks capture leaves your machine for the provider you pick, and CMEM Pro adds a sync of its own.

• Hooks: they fire at session start, on each prompt, before and after each tool call, and when a session stops or ends
• Local data: the database, settings and logs live under ~/.claude-mem
• Your provider: captured activity goes to it to be summarized
• CMEM Pro: the maker’s hosted service, pre-selected with a 30-day trial, after which memory falls back to your Anthropic plan unless you subscribe
• Cloud Sync: choosing CMEM Pro turns it on, uploading observations, session summaries and prompt text to your cmem.ai account
• Usage analytics: on by default, and npx claude-mem telemetry disable or DO_NOT_TRACK=1 turns them off
• Private text: anything inside <private> tags stays out of storage
• Maker: an independent project, not made by Anthropic
• License: Apache-2.0, though the hosted cloud services, team sync and enterprise features are not part of that release