Harness Memory

• Search first, save last

  Agents search notes at the start of a task and save what they learn (fixes, decisions, where things live) before they finish.

• Typed notes

  Notes have a kind: outcome, problem, fix, optimisation, location, decision or note, plus tags and the entities they are about.

• A small entity graph

  entity_link relates two entities, for example ConfigLoader located_in src/config.rs, and entity_neighbors shows the relations around one.

• Saved scripts

  Agents save a deterministic script by name so a repeated action runs without being worked out again. A teammate’s script runs only after you approve it.

• Local SQLite with full-text search

  Notes live in a local SQLite database with FTS5 search, keyed per git repository. Search is stemmed English keyword search.

• Optional vector search

  Adds notes close in meaning through one local embedder per computer. It is off until you enable it.

• Secrets redacted on write

  Titles, bodies, tags, entity names and relation labels pass a secret filter before they are stored. It is a heuristic and does not catch everything.

• Team memory

  Point every teammate’s server at one shared database so a fix one person’s agent saves is found by the next teammate’s agent.

• Claude Code

  Register with claude mcp add --scope user harness-memory -- harness-memory mcp.

• Codex

  Register with codex mcp add harness-memory -- harness-memory mcp, or in config.toml.

• OMP

  oh-my-pi

  Add an mcpServers entry in ~/.omp/agent/mcp.json, or use /mcp add.

• PI

  Recent versions read ~/.pi/agent/mcp.json. Older ones need the pi-mcp-adapter first.

• OpenCode

  Add a local entry to the mcp object of opencode.json.

• Any other MCP client

  Any MCP client can start the stdio server with harness-memory mcp.

1. # macOS and Linux

   $ curl -fsSL https://raw.githubusercontent.com/MSpiechowicz/harness-useful-memory/main/install.sh | sh

2. # Windows (PowerShell)

   PS> powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/MSpiechowicz/harness-useful-memory/main/install.ps1 | iex"

Prefer a download? Every release and its notes are on the Harness Memory releases page.

• Local by default. Without a team database the server opens no network connection, except that the embedder downloads its search model once when you turn vector search on.
• Notes live in one SQLite file, memory.sqlite3, in the harness-memory folder of your user data directory. The file is not encrypted.
• Secrets are redacted before anything is stored. The filter is a heuristic, so do not rely on it.
• Notes are untrusted data written by agents. Results say where each note came from, and agents are told to check notes against the code.
• Scripts run as you. A machine runs only scripts saved on it or approved by you there.

Harness Memory is the shared project memory in Harness Useful, a family of four open-source, local tools for coding-agent harnesses. Harness Desktop is the window, Harness Skills the workflow, Harness Memory the shared project memory and Harness Dashboard the usage and cost meter.

Each tool works on its own. Harness Desktop’s Recommended setup can install the others after you confirm each part and register their MCP servers in each detected harness.

• Harness Desktop Your coding-agent CLI, in a native desktop window.
• Harness Skills From a natural-language request to reviewed, verified work.
• Harness Dashboard Token usage across your AI coding tools, in one local app.

See the Harness Useful home page, all four tools side by side or the family FAQ.