DeepChat

This is a community package of DeepChat, maintained independently and unaffiliated with DeepChat or ThinkInAI. It repackages the official upstream Linux build unmodified.

DeepChat is a local-first desktop client for large language models that goes past plain chat: it hosts Model Context Protocol servers, runs skills and agents, and speaks the Agent Client Protocol, so one window can drive many providers and many tools.

• Bring your own provider — the major hosted APIs, OpenAI-compatible endpoints, or a local Ollama or LM Studio.
• Model Context Protocol support for third-party tools, with authentication and elicitation handled in the UI.
• Skills and slash commands, built-in fetch, file and shell tools, and an installer agent that configures MCP servers for you.
• Conversation search, drafts, and a local vector store for retrieval over your own material.

MCP runtime: stdio MCP servers need a Node and uv runtime, which the app downloads on first use — with the versions and checksums it pins itself — into its own data directory inside the sandbox. That step needs network access; nothing is taken from the host.

Sandbox: display, GPU, network, audio, notifications, a system tray name and the Secret Service for credential storage. There is no filesystem grant — conversations and settings live in the app's own data directory, and attaching files goes through the desktop portal. To let the built-in file tools work on a directory of yours, grant it explicitly:

flatpak override --user --filesystem=~/Projects xyz.thinkinai.DeepChat

Replace the path with your own, repeat the flag for more directories, and undo everything with:

flatpak override --user --reset xyz.thinkinai.DeepChat