Map

msgvault

Toolbox toolboxgoemailarchivalsearchduckdbsqlitemcplocal-firstwatchlistkenn-softwareGoMIT ↳ show in map Markdown
title
msgvault
type
toolbox
summary
Offline email archive with SQLite full-text, DuckDB analytics and an MCP server over it
tags
go, email, archival, search, duckdb, sqlite, mcp, local-first, watchlist, kenn-software
repo
https://github.com/kenn-io/msgvault
language
Go
license
MIT
created
2026-05-22
updated
2026-05-22

Email (and chat) archival tool: download a complete local copy of your message history, then search and run analytics over it in milliseconds, entirely offline. The pitch is "archive a lifetime of email." Part of the kenn-software-suite; the most-starred member (1.8k) but explicitly alpha.

What it does

Pulls mail from multiple sources into one local database — full Gmail backup with labels and metadata, IMAP sync from any standard server, plus MBOX and Apple Mail .emlx import. Sync is incremental (Gmail History API, so unchanged messages aren't re-downloaded) and resumable after interruption. Multi-account in a single database, with Google Workspace service-account support and domain-wide delegation for org-scale archival.

Once archived, you query offline: an interactive TUI with drill-down analytics, Gmail-like FTS5 query syntax, and semantic vector search via local embeddings (Ollama, llama.cpp, LM Studio). Hybrid search combines BM25 and vector. An MCP server exposes the archive to Claude Desktop and other agents.

How it works technically

Go 1.25+ (~99%). Two storage engines side by side: SQLite holds messages (raw MIME preserved, attachments content-addressed and deduplicated by SHA-256), and DuckDB drives the millisecond aggregate queries, with Parquet for analytics caching. That split is the design point — SQLite for transactional message store and FTS, DuckDB for the OLAP-style "how many emails per month from X over ten years" queries that a row store would crawl on.

Optional daemon mode adds an HTTP API and cron-scheduled sync. Config is TOML; MSGVAULT_HOME overrides the data location; multiple OAuth applications are supported for Workspace orgs.

Basic usage

msgvault init-db
msgvault add-account you@gmail.com
msgvault sync-full you@gmail.com --limit 100
msgvault tui

msgvault search "invoice 2023" --mode vector   # or --mode hybrid

Distribution: bash installer (macOS/Linux), PowerShell (Windows), Conda-Forge, Docker, or build from source. Building needs Go 1.25+ and a C/C++ compiler for CGO and DuckDB static linking.

Limitations

  • Alpha software. The README is explicit: APIs, storage format, and CLI flags may change without notice — back up your data.
  • DuckDB static linking means a C/C++ toolchain in the build path (not pure-Go like its sibling middleman).

Watchlist

v0.14.x and self-described alpha with a storage-format-may-change warning. Tracked on watchlist for a re-check once the format stabilizes and the alpha label drops.

  • wacli — WhatsApp CLI that syncs messages locally for offline search; same "own your message history offline" instinct, different source.
  • llama.cpp — one of the local embedding backends msgvault's vector search can use.
  • kenn-software-suite — the rest of the suite. msgvault is the personal-data member, the others target agent workflows.

Repo

github.com/kenn-io/msgvault — Go (1.25+), MIT, 1.8k stars, v0.14.1 (alpha, 27 releases). msgvault.io.

Linked from