DevOS runs 100% on your machine — your data never leaves.
31 agents. 18 tools. 10 LLM providers. 6 channels. Computer control.
Free with Ollama. Auto-detects your hardware. Zero telemetry. Zero compromise.
The foundation for autonomous planning, verified execution, and persistent learning. Built from scratch — no wrappers.
TruthCheck verifies postconditions after every action. FaultEngine classifies errors into 8 categories and applies targeted repairs. Zero silent failures.
Full hierarchy execution. Parallel tasks per project. Pause, resume, replan mid-run. Smart retry with a different strategy on each failure.
All 31 agents read and write to a shared canvas. CEO always has full mission context. No direct agent messages — canvas only.
HOT (RAM) → WARM (SQLite) → COLD (compressed). MemoryStrategy tracks successful action sequences per goal — replays them instantly next time.
First boot scans your hardware, checks installed models, and recommends the best setup for your specific machine. Never installs what you already have.
Community skills run in isolated Docker containers (512MB, 0.5 CPU, no network). Browser tasks stream via noVNC — watch DevOS browse live.
Every action runs through a unified Executor — timeout protection, exponential backoff retries, automatic fallback actions, TruthCheck on every success.
LLM plans specific line ranges to change. Never rewrites whole files. TypeScript verification after every edit. Lines changed vs preserved logged.
Ollama free by default. ModelRouter auto-picks the best model for each task type based on your VRAM. DataGuard routes sensitive data to local only.
Everything DevOS can do autonomously — no plugins, no setup, works out of the box.
VisionLoop takes a screenshot, asks the vision model what to do next, executes the action, verifies it worked, and repeats. All locally. All safely.
Give DevOS a goal like "reply to my unread emails" and watch it work. It checks for a Gmail API first, uses it if available, falls back to browser automation if not. Every action TruthCheck verified.
DevOS detects your GPU, checks what models you already have, and asks before downloading anything. Works on any hardware — RTX 4090, M3 Pro, GTX 1060, integrated Intel.
Chat like a human. Every agent action streams live on the right. Full transparency — no black box. Fully local.
Send goals from your phone. Get results on any device. 6 channels working out of the box.
Each with 400–600 word specialist system prompts. CEO orchestrates all via MissionCanvas. AgentDen gives each private workspace.
Activate once and they run forever. You check your dashboard — they do the work.
Honest benchmarks across 6 categories. Scores reflect real capability — where others win, we say so.
Public beta launching soon. Join the waitlist and be among the first developers to run a fully autonomous AI OS — with computer control — on your own machine.
Submit a feature request. Upvote what others are asking for. We read every single one.
Pick a category and describe your idea briefly.