AttogradDB · v1.1.0
Light speed memory for your agents
A local-first vector store in one SQLite file. No server, no index to tune. Search one project or one session without the rest bleeding in — and give Claude Code, Cursor, or Codex durable project memory over MCP.
Python usage Claude Code & Cursor Every other tool GitHub
Install
pip install attogradDB
Add the MCP decision-memory server:
pip install "attogradDB[mcp]"
Host plugins bootstrap that extra for you with uvx — install uv and skip the manual pip.
Where to go next
| You want to… | Go to |
|---|---|
| Give Claude Code or Cursor project memory (plugin) | Claude Code and Cursor |
| Set up Codex, Gemini CLI, OpenCode, or any MCP host | Every other tool |
| Ingest documents and search them from Python | Python usage |
| Get AttogradDB listed in a plugin marketplace | Marketplace listings |
Two tools, one database
The agent surface is deliberately small. Both tools operate on a per-project .attograd-memory.db, and the model never supplies a project or session name — the host passes the active workspace.
recall_decisions— every active decision for this project, in order. Call it before planning or editing.remember_decision— record a durable choice, with an explicitsupersedeswhen it replaces an older one.
Install the plugin in Cursor or Claude Code, or run attograddb setup for every other tool. Either way you get the same tools and the same database layout.