MemCoT skill
Long-context memory: a background service searches conversation history and a skill helps OpenClaw agents.
- 117 ★ on the repository
- Last commit
Extended access
Why this level
- The service runs in the background and processes conversation history that may contain personal data.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/memcotDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The service runs in the background and processes conversation history that may contain personal data.
This is third-party code. Review the repository files before installing.
What it does
MemCoT, an implementation of the arXiv 2604.08216 paper on test-time scaling through memory-driven chain-of-thought, searches long conversation histories for answers. The repository contains a command-line client `memcot_cil.py` with a server part: a background process keeps the built index and answers queries fast. Client commands: `start` and `stop` for the service, `status`, `session` listing sessions and index state, `add --idx N` to build a session index, `switch --idx N` to select, `search -q "<query>" -o "<dir>"` to search and answer, and `logshow` to follow the log. The `openclaw-memcot-cli` skill teaches an agent to turn user requests into these commands and answer from the retrieved context. Setup is a clone plus a Conda environment.
Who it is for. Researchers of agent memory and OpenClaw users who need search across a long conversation history.
Good fit when
- A dialog history does not fit the context and relevant fragments must be pulled out.
- A session index is built once and questions follow quickly.
- The paper's method is reproduced and compared with other memory approaches.
- An OpenClaw agent answers from retrieved context rather than guesses.
Not a fit when
- Production-ready memory with support and updates is required.
- History contains data that must not be processed by external models.
- Plain text search without embeddings is enough.
Example request
Start the MemCoT service, build the index for session 0 and find in the history what we decided about the database schema.Limitations
The repository is research code accompanying a paper; no license is set in GitHub metadata, so confirm reuse terms with the author. Installation needs a Python 3.10 environment and the dependencies from `re.txt`, including the service libraries. Building a session index uses embeddings, and model settings are described in code and configuration. The skill targets OpenClaw and a repository clone in the working directory, and is published on ClawHub as a text-only package without code. The last commit dates from October 2026.
How to disable. Stop the service with `python memcot_cil.py stop`, delete the Conda environment and the clone. Indexes and the `memcot_daemon.log` log sit beside the project.
FAQ
Which client commands exist?
`start`, `stop`, `status`, `session`, `add --idx N`, `switch --idx N`, `search -q "<query>" -o "<dir>"` and `logshow`.
Where is the service log?
In `memcot_daemon.log`. The `logshow` command follows it in real time.
What must happen before searching?
Select a session with `switch --idx N` and build its index once with `add --idx N`.
Where is the method described?
In the arXiv paper 2604.08216, linked at the top of the README.
Does it work with agents other than OpenClaw?
The command-line client is generic, but the ready skill is written for OpenClaw. Other agents need the rules ported by hand.
Discussion
No questions or reviews yet.
Related
SkillsShared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
An MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Long-term memory for agents: new facts supersede old ones, every answer traces to its source, and recall makes no LLM calls
A memory skill that stores turns, facts and episodes, verifies recall and refreshes stale entries, with an OpenClaw plugin.
README in short
The README separates installation and client use, then shows a full Python example: start the service, build an index, search and stop. A dedicated section is for agents.