cass-memory CLI
Procedural memory for coding agents: the cm CLI turns sessions into a shared playbook and serves it on request or over MCP.
- 461 ★ on the repository
- Last commit
Extended access
Why this level
- The utility reads agent session logs and writes a playbook and diary into the home directory.
- The `cm serve` server opens memory over HTTP on a local port.
Install
- Works in
Manual install
brew install dicklesworthstone/tap/cmHomebrew for macOS and Linux; the `install.sh --easy-mode --verify` script and Scoop for Windows are described in the README.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The utility reads agent session logs and writes a playbook and diary into the home directory.
- The `cm serve` server opens memory over HTTP on a local port.
This is third-party code. Review the repository files before installing.
What it does
cass-memory turns scattered agent sessions into long-term memory shared by all: every agent learns from the experience of the others. Before a task an agent requests context: `cm context "implement auth rate limiting" --json`. Always use the `--json` flag: data goes to stdout, diagnostics to stderr and exit code 0 means success. The playbook is filled through `cm onboard`, and saved rules get scores. Commands include `cm audit`, `cm diary`, `cm doctor`, `cm forget`, `cm guard`, `cm hook`, `cm mark`, `cm outcome`, `cm playbook` and `cm privacy`. A safety system called Trauma Guard keeps the agent from repeating mistakes that already cost data. `cm serve` starts an MCP server over HTTP (127.0.0.1:8765 by default). Installation uses the `install.sh` script, Homebrew (`brew install dicklesworthstone/tap/cm`) or Scoop on Windows.
Who it is for. Developers who run several coding agents in one project and want discovered techniques and mistakes to survive between sessions.
Good fit when
- Before a task the agent needs rules and lessons from past project sessions.
- An agent steps on the same rake a second time and it must be locked in as a rule.
- The experience of Claude Code and Codex must be merged into a shared playbook.
- An agent must receive memory over the network through an MCP server on a local port.
Not a fit when
- Memory is needed for one short project and a playbook is not.
- Session history contains sensitive data that must not be processed.
- Memory is wanted with no separate `cass` index.
Example request
Before starting the task run `cm context "implement auth rate limiting" --json` and apply the rules it finds.Limitations
Memory is built from agent session logs, so conversation history is processed on your machine and `cm privacy` sets privacy options. Initial playbook filling goes through `cm onboard` and takes time. Add `cm context` calls to agent instructions or hooks, because the utility does not do this itself. The `cm serve` MCP server listens on 127.0.0.1 by default. The last commit in the repository is from November 2026.
How to disable. Remove the `cm` binary (`brew uninstall dicklesworthstone/tap/cm`) and the playbook directory with the diary, and drop `cm context` calls from agent instructions.
FAQ
How do I install cass-memory?
With the README `install.sh --easy-mode --verify` script, Homebrew `brew install dicklesworthstone/tap/cm` or Scoop on Windows. The command is called `cm`.
How does an agent get memory?
With `cm context "task" --json` before work starts. For network use there is `cm serve`, an MCP server over HTTP at 127.0.0.1:8765 by default.
Why the --json flag?
In agent mode data goes to stdout, diagnostics to stderr and exit code 0 means success. That keeps output parseable by a program.
How do I fill the playbook?
With `cm onboard status --json`, `cm onboard sample --fill-gaps --json` and `cm onboard read <session file> --template --json`. `cm onboard mark-done` marks a processed session.
What does Trauma Guard do?
A safety system that keeps an agent from repeating dangerous actions that already caused data loss before. It is configured with `cm guard`.
Discussion
No questions or reviews yet.
Related
CLI toolsmemory.cpp compiles repository history into compact agent context and filters out stale items, secrets and contradictions
An MCP server that gives an agent memory across sessions, projects and machines: decisions, dead ends, preferences and skills
Shared memory for coding agents: messages in SQLite, old context compacted into a summary tree and MCP tools to search it.
A local memory MCP server for coding agents: records issues, attempts, fixes and decisions and warns before repeating a failed approach
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
README in short
Memory that all your agents share: session experience becomes a playbook with scores. Served through a JSON CLI or over MCP.