YantrikDB CLI
Agent memory in Rust: forgets the unimportant, merges similar items, flags contradictions, runs locally, over HTTP and via MCP.
- 179 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The server stores agent memories, including decisions and project details, in a local file or cluster.
- For network use the database token and server address are passed in the client configuration.
Install
- Works in
Manual install
pip install yantrikdb-mcpThe MCP server from the neighboring repository; then a `yantrikdb` block with the command `yantrikdb-mcp` in the client config.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server stores agent memories, including decisions and project details, in a local file or cluster.
- For network use the database token and server address are passed in the client configuration.
This is third-party code. Review the repository files before installing.
What it does
YantrikDB is a memory database for agents that manages what it stores instead of only keeping it. After tens of thousands of memories, recall quality drops: items pile up without consolidation, forgetting or conflict resolution. YantrikDB adds all three. Each memory has an importance and a half-life, so after a week an unimportant item stops surfacing. Similar items merge and contradictions are flagged. The engine embeds in Python, runs as a Docker server (`docker run -p 7438:7438 ghcr.io/yantrikos/yantrikdb:latest`) or connects through MCP. The shortest path for Claude Code, Cursor and Windsurf is `pip install yantrikdb-mcp` from the neighboring repository and one config block. The agent then gets 15 memory tools, including `remember`, `recall`, `forget`, `correct` and `think`, and a local SQLite file `~/.yantrikdb/memory.db`. For shared memory across machines the MCP can point at a YantrikDB cluster.
Who it is for. Agent developers and Claude Code users whose file memory such as CLAUDE.md has grown past what fits in context.
Good fit when
- An agent notes file has grown to thousands of lines and answer accuracy is dropping.
- The agent should forget urgent chores after a day and remember decisions for months.
- Two notes about the same project contradict each other and the agent must know.
- A team wants shared agent memory on a cluster instead of local files.
Not a fit when
- A short `CLAUDE.md` is enough and memory is not growing.
- You need an ordinary vector database for document search.
- You need a stable product: some project materials are still evolving.
Example request
Remember that the project uses PostgreSQL 16, and on the next database question recall it and tell me if something different was recorded earlier.Limitations
The `yantrikdb-mcp` MCP server lives in a separate repository and installs through `pip`, while this repository holds the Rust server itself. The authors published a correction notice: phase 3 write-ups used a Python simulator instead of the engine, with a rerun on the real engine in `docs/phase3e/`. Token savings figures from the README should be checked on your own data. A cluster needs `YANTRIKDB_SERVER_URL` and a database token `YANTRIKDB_TOKEN`. The project is part of a broader YantrikOS plan, so interfaces may change.
How to disable. Remove the `yantrikdb` entry from the MCP config and the package with `pip uninstall yantrikdb-mcp`. Delete the `~/.yantrikdb/memory.db` memory file by hand if you no longer need it.
FAQ
What is the fastest way to try it?
With `pip install yantrikdb-mcp` and one MCP config block. No environment variables are needed and memory lives in local SQLite.
How do I run the server?
With `docker run -p 7438:7438 ghcr.io/yantrikos/yantrikdb:latest`. The Python client installs with `pip install yantrikdb`.
How does forgetting work?
A record gets `importance` and `half_life`. Relevance decays and after several periods the record stops surfacing unless queried explicitly.
Which tools does the agent get?
Fifteen, including `remember`, `recall`, `forget`, `correct` and `think`, which consolidates items and detects conflicts.
Can I connect over HTTP or SSE?
Yes. Run `yantrikdb-mcp --transport sse --port 8420` and set the `/sse` URL with an `Authorization` header in the client.
Discussion
No questions or reviews yet.
Related
CLI toolsMenteDB keeps agent memory in a Rust database: an MCP server for Claude Code and Cursor in cloud or self-hosted mode.
A long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
AgentDB keeps agent memory in an .rvf file and, with 41 MCP tools, recalls past solutions while learning from feedback
Local memory for agents in a single Rust binary: stores, searches by meaning and keyword, works with no network and no keys
Local memory for agents with a review queue: SQLite, an audit log and a token ledger, one vault for Claude Code, Codex and Cursor
Memory for agents over MCP: five tools, one local SQLite file and layers modeled on human memory, no account
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| YantrikDB (this card) | no | 179 | ||
| Agent Memory Engine | no | not required | 127 | also memory for agents; here half-life forgetting, merging of items and contradiction search |
| Basic Memory | no | not required | 4,557 | keeps memory as Markdown notes; here a database with importance scoring, decay and a cluster |
README in short
The project describes a cognitive memory engine: forgetting, consolidation and contradiction detection. It gives a quick start for Claude Code, Cursor and Windsurf, a list of MCP tools and Python examples. A table compares token savings and a correction notice explains which measurements need rerunning.