DecisionsAI is a free desktop AI agent for developers who want JARVIS on their machine. Telegram remote control. Ticket Boards. Step Runner workflows. Actions and Snippets you trigger by voice. Vision and computer use. Ollama integration for swappable local models. Google Gemini for multimodal reasoning. Kilo gateway for 500+ models. Drop files on the Oracle for instant context. Swap every component — LLMs, TTS, STT, API keys — without touching code. Open source. macOS, Windows, Linux.
Type "remote" in Telegram and you get a live stream of your desktop on your phone. Click, type, drag files — all from the beach, the coffee shop, or the other room. Your AI agent runs 24/7. You don't have to be at your desk.
Built-in Ticket Boards with AI agent automation. Assign agents to boards — they process tickets on schedule, move cards between columns, and report back. Like having a junior developer who never sleeps. Create multiple boards for different projects.
Chain multi-step instructions into automated workflows. "Open Chrome, go to GitHub, check PRs, summarize, and send to Telegram." Run once or schedule to repeat. Like terminal commands in your IDE, but powered by AI and triggered by voice.
Record clicks, typing, and scrolling — then replay by voice or on schedule. Add trigger words so the agent knows when to run them. "Good morning" opens Slack, checks email, and starts your music. Never do the same thing twice.
Save your most-used text and commands. Say "deploy" and the agent stages, commits, pushes, and notifies. Say "standup" and it summarizes yesterday from git log and your Ticket Board. No more context switching.
Take a screenshot and ask "what's wrong?" The agent sees your screen, reads errors, and suggests fixes. Computer use means it can click, type, and navigate — not just talk about what to do, but actually do it. Supports Anthropic Opus 4.7 and Google Gemini for vision and computer use.
Run Ollama locally for free, unlimited inference. Use Ollama Cloud for flat-rate subscriptions. Switch to OpenAI GPT-5.2, Anthropic Opus 4.7, Google Gemini, or route through Kilo and OpenRouter for access to 500+ models. Add your own API keys in Third Party settings. Groq for ultra-fast inference. All from Settings, no code required.
We love Ollama. New models drop and you swap them in seconds from Settings. Llama, Mistral, Qwen, CodeLlama — separate models for conversation, coding, and vision. Download directly from the UI. No terminal wrangling.
Google Gemini brings powerful multimodal reasoning to DecisionsAI. Use Gemini for vision tasks, long-context understanding, and complex reasoning — all configurable from Settings alongside your other LLM providers.
Kilo (kilo.ai) is a universal AI inference gateway that routes your requests to 500+ models from Anthropic, OpenAI, Google, Mistral, and more. Bring your own keys or use Kilo credits. One API endpoint, every model. Integrated directly into DecisionsAI Settings.
Add your OpenAI, Anthropic, Google, Groq, Kilo, and ElevenLabs API keys in Third Party settings. Enable what you need. The agent routes to the right provider based on your LLM settings. Cloud when you want it, local when you don't.
Seven skins: Oracle, Clippy, Hayley, Nugget, Rusty, Masko, and Madame Patate. Each with custom idle, thinking, and working animations. Drag and drop files onto it for instant context — PDFs, images, audio, code. Right-click for quick actions. It lives on your desktop so you never have to switch windows.
Chat, Ticket Boards, Step Runner, Projects, Snippets, Actions, and Settings — all in your browser at localhost:8765. Manage everything from one interface. Real-time logs showing every tool call, STT pickup, and TTS playback.
Link local folders to give the agent deep context about your codebase. Each project gets its own RAG index, startup instructions, and Ticket Board. Switch projects and the agent instantly knows your code.
You don't switch between windows. The Oracle floats on your desktop. You talk, it does. You drag files, it reads them. You say "remote" on Telegram and your phone becomes your desk. Every component swaps out — LLMs, speech models, voice engines, API keys. It's your machine, your rules, your AI.
Pipecat — real-time voice orchestration. Ollama — local model inference. Google Gemini — multimodal reasoning. Kilo — universal AI inference gateway for 500+ models. OpenRouter — route to hundreds of models. Whisper.cpp — offline speech recognition. Vosk — ultra-low latency STT. Kokoro TTS — local voice synthesis. Qwen3-TTS — reference-audio voice cloning. ElevenLabs — premium cloud TTS with voice cloning. OpenAI — GPT vision and tool calling. Anthropic — Opus deep reasoning and computer use. Groq — ultra-fast inference. Playwright — browser automation. PyAutoGUI — desktop control. LlamaIndex — RAG indexing. PyQt6 — desktop UI. Masko — custom skin.
One command installs everything. Creates ~/.decisionsai/, clones the repo, runs setup. Available for macOS, Windows, and Linux. Free and open source on GitHub.
Yes — actively shipping. DecisionsAI has been an ongoing experiment in AI-assisted development itself. We started with Cursor, moved through Kiro and Claude Code, and now build primarily with pi running GLM-5.1 — and the cost effectiveness is remarkable. That journey shaped the product: we eat our own dog food, iterating with the same AI tools we integrate into the agent. The way we develop now — AI-first, voice-driven, remote-capable — is exactly the workflow DecisionsAI gives you.