Kept

Product Introduction

1. Definition: Kept is a local-first AI chat archiver and productivity tool. It is a desktop application and browser extension that captures, indexes, and stores conversations from major AI platforms (ChatGPT, Claude, Gemini, Grok, Kimi) as plain text, Obsidian-compatible markdown files on your local filesystem.
2. Core Value Proposition: Kept exists to provide users with complete, private ownership and control over their AI chat histories. It solves the problem of fragmented, platform-locked conversations by creating a unified, searchable, and portable knowledge vault on your own machine, with no cloud dependency for core functionality.

Main Features

1. Local-First Markdown Vault: Every AI conversation is saved as a single, plain markdown file with YAML frontmatter (containing metadata like platform, date, and tags) to a designated directory (~/.kept/vault/ by default). This ensures conversations are human-readable, grep-able, and compatible with any text editor or note-taking system like Obsidian.
2. Full-Text & Semantic Search: Kept uses SQLite's FTS5 engine to index every word in the archive, enabling instant, filterable full-text search. Its "Smart Recall" feature adds a semantic layer, allowing users to ask questions in plain English and receive relevant passages with citations, powered by user-configured AI APIs (OpenAI, Anthropic, etc.) or local models via Ollama.
3. MCP (Model Context Protocol) Server Integration: Kept ships with kept-vault-server, an MCP server that exposes the local vault for read, write, and search operations. This allows any MCP-aware client (Claude Desktop, Cursor, Cline) to directly integrate with and query the user's archived chat history, bridging personal knowledge with development environments.
4. Knowledge Graph & Auto-Tagging: The tool automatically extracts entities, decisions, and relationships from conversations to build a local knowledge graph. It also clusters conversations by theme without manual tagging, enabling users to visually explore and discover connections across different chats and AI platforms.
5. Bring Your Own Keys (BYOK) & Local Continuation: Users can configure their own API keys for OpenAI, Anthropic, or OpenRouter within Kept. This allows them to "continue" any archived conversation directly within the Kept app using any supported provider, decoupling ongoing work from the original chat platform.

Problems Solved

1. Pain Point: Vendor lock-in and data portability in AI chat platforms. Users cannot easily export, search across, or maintain a unified history of conversations from different AI providers (ChatGPT, Claude, etc.), leading to lost context and fragmented knowledge.
2. Target Audience: Power users of generative AI, including software engineers, researchers, technical writers, and knowledge workers who rely on multiple LLMs for coding assistance, research, content creation, and decision logging. It is particularly valuable for Obsidian users and developers using MCP-enabled tools.
3. Use Cases: A developer researching a migration across several weeks using both ChatGPT for code examples and Claude for architectural advice can have all conversations archived, linked, and fully searchable in one place. A team can use the local vault and MCP server to share a common knowledge base within development tools like Cursor. An individual can maintain a permanent, private decision log from AI consultations.

Unique Advantages

1. Differentiation: Unlike cloud-based note-taking apps or manual copy-paste methods, Kept is architected from the ground up as a local-first, privacy-focused tool. It differs from simple browser history exporters by offering deep indexing, semantic recall, a knowledge graph, and deep integrations (MCP, Obsidian) that transform a raw archive into an active knowledge base.
2. Key Innovation: The combination of a local markdown vault with a built-in MCP server is a significant technical innovation. It treats the user's personal chat archive as a first-class data source for the AI tooling ecosystem, enabling bidirectional flow between personal memory and AI-assisted development environments without ever sending data to a third-party cloud service by default.

Frequently Asked Questions (FAQ)

1. Is Kept completely free and open source? Yes, Kept is released under the permissive MIT license on GitHub. The public repository contains the full feature set; there is no premium tier, freemium model, or gated functionality, aligning with its ethos of user ownership and transparency.
2. Does Kept send my chat data to its own servers? No. By default, no chat data leaves your machine. Conversations are stored locally in SQLite and as markdown files. The only exception is if you opt to use cloud-based features like "Smart Recall" or the knowledge graph with a cloud API (e.g., OpenAI), in which case your search queries are sent to that provider, similar to a standard API call.
3. How does Kept's Obsidian export work? Kept maintains its primary vault in ~/.kept/vault. The Obsidian export is an optional, one-way sync operation that copies this vault into an existing Obsidian folder. The reconciliation logic is designed to preserve any edits made on the Obsidian side, preventing overwrites during subsequent exports. It is not a real-time, two-way synchronization system.
4. Can I restore an archived conversation back into ChatGPT or Claude's web interface? No, Kept does not currently support pushing conversations back into vendor web UIs. The platforms do not offer public APIs for writing to chat history, and Kept avoids unreliable DOM manipulation. Instead, you can "continue" any chat within the Kept app using your own configured API keys for any supported provider.
5. What are the system requirements for running Kept? Kept is a cross-platform desktop application available as a standalone executable for Windows (.exe), macOS (.dmg), and Linux (.AppImage, .deb). It requires a modern operating system and the ability to sideload a Chrome extension (for the capture functionality). The local indexing and search are efficient and designed to handle archives containing tens of thousands of conversations.