🎉September 6, 2026 — the first public release of AI Assistant is live! I am actively developing it and fixing bugs, and I truly hope you will enjoy my vision for the app. Thank you for being here!
Your personal AI assistant for chats and projects
Cross-platform desktop app for chatting and working with any projects. Provider-agnostic: you configure your own LLM endpoints (Ollama, OpenAI-compatible, Anthropic, OpenRouter, llama.cpp, vllm, …), system prompt, tools, MCP servers, and external code agents (cline / opencode / claude-code, etc.). No login or sign-up. No required online accounts. No payments or paywalls. No tracking. No brands. Just you and your usual useful tools.
Native builds for Windows, macOS and Linux.
Take a look inside
Screenshots from the desktop app — click any image to enlarge.
Capabilities
Every feature here exists because real daily work needed it — provider-agnostic, project-aware, and ready to extend with agents, MCP, and your own tools.
Chats & Projects
Standalone chats
Quick, no-strings chats for asking a question, brainstorming, or getting something done — no project or folder attached. Just a conversation.
Projects with shared context
Projects go further: chats inside a project share a common context that the model can draw on in every new chat.
Project directory
Bind a project folder and the model gets direct access to read files and do work inside it.
Per-project system prompt
Each project can override the app's default system prompt to shape the model's overall behavior for that project.
Move chats freely
Pull a chat out of a project into a standalone chat, or drop a standalone chat into a project so it becomes part of the shared context.
Project rules
Custom rules are small, separate system prompts that guide the model's behavior within one specific project.
Kanban task board
Every project has a Kanban board — Backlog, To Do, In Progress, Review, Done — and tasks can be launched into execution right from the board.
Drag-and-drop
Reorder tasks by dragging them on the Kanban board or in the list view, whichever you prefer.
RAG (optional)
Projects can opt into retrieval-augmented generation through an embedding model, so the model can pull in relevant context from your files.
Providers
As many providers as you want
Add local and remote providers alike, and pick a model from any of them directly in a chat.
Default, fast, and embedding models
Set a default model for everyday chats, an optional fast model for quick tasks, and an embedding model for RAG.
Agents & MCP
External agents
Add as many external agents as you like — opencode, claude-code, cline, ollama-agent, or any other. AI Assistant does a lot on its own, but with external agents it becomes an orchestrator that delegates and coordinates their work.
MCP servers
Third-party MCP servers extend AI Assistant with whatever capabilities you need — just plug them in.
MCP Web UI
An MCP server running locally can expose its own Web UI, which you can pin as a button on the top bar for quick access — like the Codebase Memory Web UI shown in the screenshots, opened in its own window right inside AI Assistant.
Built-in tools
Built-in tools
The built-in toolset covers most of the tasks an agent needs day to day.
Web search
A built-in web-search tool lets you choose exactly which search engines it uses.
Webhooks for models
Preconfigure webhooks and the model can call them on demand with data. For example, set up a webhook that posts text to a Slack channel, then ask the model to do the research and send the result to Slack.
Custom skills
Define your own skills and add them manually to chats whenever you need them.
Prompts
Reusable prompts
Save as many prompts as you want and pin the ones you use most to the home screen to launch them instantly. For example, build a prompt that collects the week's completed work and sends it to Slack — then run it every Friday without retyping.
Why AI Assistant
Minimal defaults, full user control. Everything is configurable and overridable.
Bring your own provider
Connect any LLM endpoint — local Ollama or any remote API. OpenAI-compatible and Anthropic providers with streaming, tool use, thinking blocks, and vision.
Private by design
API keys live in the OS keychain, never in plaintext. Your data stays on your machine. No login, no tracking, no brands.
Chats & projects
Standalone chats and Projects — sets of related chats with shared context and connected directories.
External code agents
Bridge to cline, opencode, claude-code and others via PTY — with run monitoring, diff view, and approve/reject workflow.
MCP & RAG
Model Context Protocol client with tool discovery, plus per-project retrieval-augmented generation with embeddings and vector store.
Multilingual
Full interface in 11 languages. Switch instantly without losing context.
Prompt library
Saved prompts with favorites, per-chat system prompts, and per-project rules.
Project task board
Kanban and list views for project tasks; tasks can be launched as agent runs.
Activity stats
Token usage, message counts, and timing metrics with a daily activity graph.