DataMind plugin
DataMind gives Codex a shared data layer: knowledge base, SQL, graph and memory, with a receipt for every write.
- 455 ★ on the repository
- Last commit
Extended access
Why this level
- The plugin reads document folders and writes to the knowledge, graph and memory stores.
- Embeddings are sent to the chosen provider together with document text.
Install
- Works in
Manual install
python -m pip install -e .
./scripts/install_codex_plugin.shRun the commands inside the repository clone. A standalone MCP server ships in the package and starts as `python -m datamind.mcp`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin reads document folders and writes to the knowledge, graph and memory stores.
- Embeddings are sent to the chosen provider together with document text.
This is third-party code. Review the repository files before installing.
What it does
The datamind-context plugin connects Codex over MCP to a shared workspace data layer. Codex does the planning, reads the documents and writes the answers; DataMind stores the data, runs the tools and returns evidence and write receipts. The layer has five parts: a knowledge base with semantic search, SQL for exact records and aggregation, a graph of typed entities and relations, versioned procedures (skills) and memory of user facts and preferences. A folder of documents is ingested as a workspace: DataMind checks source versions by SHA-256 and publishes the result as one complete revision. A failed publication preserves the previous head, and a reader stays pinned to one revision for the whole operation.
Who it is for. Codex users and teams who want shared memory over a document folder and a database, with repeatable answers and a trail of what was written.
Good fit when
- A folder of specs and notes should become a base Codex answers from with source references.
- Relations between companies must be extracted from contracts and traversed as a graph.
- Several Codex sessions read the same data and must not see a half-written state.
- One document changed and only it should be updated without rebuilding everything.
Not a fit when
- One quick search over a few files is enough: ordinary repository search will do.
- The data is confidential and document text cannot go to an external embedding provider.
- The work happens in Claude Code or Cursor: the plugin and SDK are built for Codex.
Example request
Ingest the docs folder into DataMind, then answer who owns Project A and show the source.Limitations
The plugin ships for Codex. Installation needs Python and a cloned repository: first `python -m pip install -e .`, then the plugin install script. The knowledge base needs a separate embedding provider with its own key. The local NetworkX and JSON graph supports a single writer; distributed storage remains separate work. Passing tests confirm the stated contracts, not the correctness of every model answer.
How to disable. Remove the DataMind Context plugin in Codex and stop the `python -m datamind.mcp` process if you started it separately. Delete the data in the DataMind directory by hand.
FAQ
How does work with the plugin begin?
Each session opens with a `datamind_operation_begin` call in write or read mode, and its `operation_id` is passed to every following tool.
Does DataMind need a language model key?
Not for the graph, SQL, skills and memory tools. Only the embedding provider behind the knowledge base needs a key.
How do I confirm a write succeeded?
Check `results[].status` in the tool response and read the receipt. Until the status is confirmed, success should not be claimed.
Is there a CLI and HTTP?
Yes. The `datamind ask`, `datamind store` and `datamind chat` commands and the HTTP API run over the same data layer and the same main Codex agent.
Which stores work locally?
SQLite, NetworkX, Chroma and LanceDB are supported locally. CI also tests against real MySQL and Dolt servers.
Discussion
No questions or reviews yet.
Related
PluginsThe Dex plugin: an agent explores a database, builds dbt models and a semantic layer and tracks schema drift under cost control
A molecular biology workbench for Codex and Claude Science that opens DNA, RNA and protein sequences, alignments and results.
DataLens Dev MCP
datalens-dev-mcp
Local MCP server and Codex plugin for Yandex DataLens dashboard development: read, author, publish, back up and clean up objects
A CLI and MCP server that extracts structured knowledge from documents: graphs, hypergraphs, and fact search with sources
Skill and CLI for tables: cleaning with recorded decisions, level-based statistics, a PDF report and a separate audit.
A local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| DataMind (this card) | no | 455 | ||
| knowledge-rag | no | not required | 305 | an MCP server for document search; DataMind adds SQL, a graph, memory and workspace revisions |
| agent-memory | no | 4,315 | stores agent memory; DataMind joins memory with a knowledge base, SQL and a graph |
README in short
Version 2.0 removed the internal store and retrieve agent loops: the main Codex agent carries all the logic while DataMind stays a data layer. The plugin, SDK, CLI, HTTP API and web UI share one store.