TOMC Memory plugin
TOMC Memory is a task-memory plugin that gives an agent the latest value of a changed fact together with its source line.
- 264 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin stores project notes and history locally, which may include sensitive code and decisions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/tomc-memoryDetects 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 plugin stores project notes and history locally, which may include sensitive code and decisions.
This is third-party code. Review the repository files before installing.
What it does
TOMC Memory stores project notes and prepares compressed context from a history the user supplies. When a fact changed several times, for example a framework moved from Flask to Django and then FastAPI, the agent receives only the latest value and the line it came from. Claude Code gets a plugin with an MCP server, Codex a separate archive with an installer, and Cursor, Claude Desktop and TRAE a setup through `tomc setup`. Notes stay local in `~/.tomc/memory.sqlite3`. The author calls version 0.3.1 a research preview.
Who it is for. Developers with long project chats in Claude Code or Codex where decisions get revised and stale versions get in the way.
Good fit when
- A project decision changed several times and the agent returns the old one.
- A chat history must be compressed before an API call while keeping current values.
- Notes need to move from Codex to Claude Code on one computer.
- Tokens go to a long history and you want to test the savings on your own tasks.
Not a fit when
- You want to shrink a chat the agent already runs: the plugin only handles history you hand it.
- The history is under about 10K tokens, where savings are small.
- A reworded update must never be missed: matching works on words, not meaning.
Example request
Try TOMC Memory: save the framework decision and return the current value with its source line.Limitations
This is a research preview. Savings figures come from the author's own tests and the paper is not public yet. Cursor, TRAE and WorkBuddy setup is written but tool calls were not verified on real machines, and Claude Desktop installation was not checked in the app. Matching works on words, so a phrasing such as "scratch that, go with Django" can be missed, mostly at small budgets. The Claude Code and Codex plugins need uv, and running from source needs Python 3.10 or newer.
How to disable. Remove the plugin through `/plugin` or the server entry from the client config. `forget_memory` deletes one notebook and `tomc erase-all --yes` deletes everything.
FAQ
How do I check the plugin works?
After installing, tell the agent "Try TOMC Memory" or run `/tomc-memory:try`; the terminal has `tomc try`. If something fails, run `tomc doctor`.
Where are notes stored?
Notes saved with `remember_memory` stay on the computer in `~/.tomc/memory.sqlite3`. Results of `prepare_context` and `recall_memory` are not saved unless `tomc settings --save-mcp-results on` is set.
How do I connect Cursor?
Run `tomc setup --client cursor --write` in an installed copy and restart the client. The command backs up the config first and keeps other servers.
How many tokens does it save?
The author reports roughly 10–20% of input on earlier previews at the default budget. Short histories gain less, so test on your own tasks.
How do I erase all data?
`tomc erase-all --yes` deletes all notes. The `forget_memory` tool deletes a single notebook.
Discussion
No questions or reviews yet.
Related
PluginsA long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
A memory plugin for Claude Code and Codex: team knowledge from wikis, tickets and chat with citations and stale-fact checks.
Local memory for Claude Code, Codex and Claude Desktop: one Markdown folder and one MCP server shared by all.
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
A local MCP memory server for coding agents: decisions, errors and facts stay on your machine, with semantic and graph recall
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| TOMC Memory (this card) | no | 264 | ||
| NeuralMemory | no | not required | 244 | stores long-term agent memory; TOMC Memory returns the latest value of a changed fact with its source line |
| Context Mode | no | not required | 34,013 | saves context on tool output; TOMC Memory compresses project history and notes |
README in short
The README opens with caveats: what the plugin does not shrink, where savings are small, what is stored and which clients are unchecked. Each figure links to an evidence page with conditions and failures.