Install

Midas is Apache-2.0, local, and free. One command gets the SDK on your machine — no accounts, no keys, no telemetry.

Requirements

• Python 3.10 or newer (CPython). The pure-Python path works on Linux, macOS, and Windows.
• Git, to install the SDK from the repository.
• ~200 MB free disk for the local embedder model on first run.
• SQLite ships with Python — no separate database process to run.
• Optional: uv if you prefer it over pip. Node 18+ if you want the TypeScript client.

No GPU, no server, no external service is required. Midas runs entirely on your machine and writes to a single SQLite file.

Install the Python SDK

pip install "midas-memory[all] @ git+https://github.com/vornicx/Midas"

The [all] extra pulls the local embedder, importance scorer, and the MCP server. Drop it to install only the core.

With uv:

uv pip install "midas-memory[all] @ git+https://github.com/vornicx/Midas"

Verify

python -c "import midas; print(midas.__version__)"

First run

from midas import Memory, LocalEmbedder

mem = Memory(embedder=LocalEmbedder())
mem.remember("Launch is on Nov 12.", kind="fact", importance=5)
print(mem.build_context("When is launch?", token_budget=128))

Midas creates ~/.midas/memory.sqlite3 on first write. Delete the file to reset state.

Install the TypeScript client (optional)

npm install midas-memory

import { MidasClient } from "midas-memory";

const mem = new MidasClient({ db: "~/.midas/memory.sqlite3" });
await mem.remember("Launch is on Nov 12.", { kind: "fact", importance: 5 });
console.log(await mem.buildContext("When is launch?", { tokenBudget: 128 }));

The TypeScript client talks to the same SQLite file the Python SDK writes.

Run the MCP server

Expose Midas to Claude Desktop, Cursor, VS Code, or any MCP-capable agent:

midas mcp serve

Point your client at stdio or the printed HTTP/SSE URL. See the README for per-client config snippets.

Upgrade

pip install --upgrade "midas-memory[all] @ git+https://github.com/vornicx/Midas"

Uninstall

pip uninstall midas-memory
rm -rf ~/.midas

Troubleshooting

• command not found: midas — ensure your Python bin directory is on PATH, or invoke via python -m midas.
• First remember() is slow — the local embedder downloads on first use, then caches under ~/.midas/models.
• Permission errors on ~/.midas — set MIDAS_HOME to a writable directory.

Next

• Introduction — what Midas is and why it exists.
• Complete overview — the memory model, guard, and MCP surface.
• Benchmarks — reproducible retrieval and cost numbers.