selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Kept logo

Self-host Kept

v1.0.1

Local-first AI chat archive with search, graph, and MCP

ReleaseOpen SourceMIT
Updated4 months ago
LanguageTypescript
Setup difficultyModerate
Stars
147
Port
18241 (Local server for browser extension communication)

What is Kept?

Kept is a local-first desktop application that archives your AI conversations from major providers as plain Markdown files with YAML frontmatter in a local vault. A Chromium browser extension captures conversations via provider APIs and sends them to the Tauri 2 desktop app running on localhost. The app indexes content with SQLite FTS5 for fast full-text search, builds a knowledge graph in CozoDB for topic and entity discovery, and supports semantic recall via configured LLM providers or local Ollama models. It includes an MCP server so agents like Claude Code, Cursor, or Cline can read, write, and search the vault. The vault is Obsidian-compatible and can be exported to an existing Obsidian vault with edit preservation. No cloud sync or account is required; all data stays on the user's machine by default. The project is MIT licensed with no premium tier.

Should you self-host Kept?

Kept scores 46/100 on selfhost.directory (how scoring works) and ranks #41 of 69 in AI Chat. It has 147 GitHub stars.

It shipped 2 stable releases in the past year. The latest, v1.0.1, came out on 12 May 2026. New versions land roughly every 8 days.

Setup is moderate: expect some configuration beyond a single command. It is MIT-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is ChatGPT Next Web (85/100, 88 905 stars). It scores higher overall, so compare the two before you pick. See Kept vs ChatGPT Next Web side by side.

Features

  • AI conversation capture
  • Auto-topic clustering
  • Background auto-sync
  • BYOK (Bring Your Own Keys)
  • Claude Code integration
  • Continue chats in-app
  • Digests generation
  • Full-text search (SQLite FTS5)
  • Knowledge Graph
  • Markdown vault
  • MCP Server
  • Obsidian export
  • PDF and image reading
  • Semantic recall

Kept README

How to configure Kept

  • config.tomlMain configuration file for the desktop app

    Main configuration file for the desktop app

  • KEPT_VAULT_PATH~/.kept/vaultPath to the Kept vault directory for the MCP server

    Path to the Kept vault directory for the MCP server

Kept releases & version history

FAQ

Can I contribute?

Yes, the repo is on GitHub under MIT. Open issues, send PRs, or star it. The desktop app is Rust + TypeScript; the extension is vanilla JS; the MCP server is TypeScript.

How does the Obsidian export relate to the markdown vault?

The markdown vault is Kept's default directory under your home folder. Obsidian export is a separate opt-in operation that copies the vault into a folder you point Kept at, preserving edits made on the Obsidian side. It's not full two-way sync.

Can I push a conversation back into ChatGPT or Claude?

Not currently. Vendors don't expose write APIs for chat history, and the project avoids fragile DOM-injection workarounds. If a clean way appears, it will be opt-in and clearly labeled.

What does MCP support mean in practice?

Kept's own agent can call any MCP server you connect, and Kept ships kept-vault-server on npm which exposes your vault over MCP so you can plug your archive into other MCP-aware clients to read, write, or search it.

Why sideload the extension instead of Chrome Web Store?

The Web Store review queue takes time, especially for an extension that touches multiple AI sites. The extension is feature-complete; loading it unpacked is straightforward. It will go through the store eventually.

Does any data leave my machine?

Nothing leaves your machine by default. Conversations are stored locally in SQLite and as Markdown. Only if you use Smart Recall or knowledge graph with cloud APIs (OpenAI, Anthropic, OpenRouter) do queries go to those providers. Using Ollama keeps everything on-device.

Is everything really in V1? No freemium?

Yes, the public release on GitHub is the full feature set with no premium tier or gated build. The project is MIT licensed.

More in Productivity

Rate this project

Click to rate — no account needed.

0 comments

Posting as guest
Loading comments…