Autonomous, Model-Agnostic AI Software Engineering Inside VS Code
COMU is an open-source AI software engineering agent that investigates codebases, creates structured implementation plans, modifies files with deterministic concurrency control, verifies builds & tests, diagnoses failures, repairs bugs, and seeks human approval for critical actions.
⚡ Key Capabilities
Bring Your Own Key (BYOK): Connect your own provider credentials directly in VS Code. First-class support for NVIDIA Nemotron 3 Ultra and local offline models via Ollama.
Interactive Implementation Plans: Live progress tracking across plan steps with real-time SSE event streaming.
Automated Verification Engine: Automatically runs linters, typechecks, build commands, and test suites to validate edits before completion.
Failure Diagnosis & Self-Repair: Pinpoints root causes and performs bounded self-repair cycles when tests or builds fail.
Controlled Git Governance: Human-in-the-loop interactive approval cards for staging, commits, and remote pushes.
Encrypted Local Storage: API keys are saved exclusively in VS Code SecretStorage and never exposed or committed to Git.
🚀 Getting Started
1. Instant 1-Click Launch (Auto-Starting Backend)
The COMU extension automatically launches and manages the local Agent Runtime backend process in the background. No manual terminal commands are needed to get started!
2. Connect Your AI Provider (BYOK)
Open the COMU Sidebar in VS Code (Ctrl+Shift+P → COMU: Open Chat).
Click the ⚙ Settings button in the header (or click Connect NVIDIA Nemotron on the onboarding card).
Under the NVIDIA card, paste your NVIDIA API Key (from build.nvidia.com).
Click Save and Test Connection to verify connectivity with latency reporting.
(Alternatively, select Ollama (Local) to run models completely offline with zero network calls).
(Optional for developers): If you want to run a custom development runtime server on port 3456, COMU will automatically detect and connect to your existing instance instead of starting a new process.
🛠️ Contributed Commands
Command
Title
Description
comu.openChat
COMU: Open Chat
Opens the COMU AI engineering sidebar
comu.openProviderSettings
COMU: Open Provider Settings
Opens the BYOK Provider configuration page
comu.configureNvidia
COMU: Configure NVIDIA Provider
Directly jumps to NVIDIA Nemotron setup
comu.testProviderConnection
COMU: Test NVIDIA Connection
Pings the NVIDIA endpoint and displays latency
⚙️ Extension Settings
Setting
Default
Description
comu.runtime.baseUrl
http://localhost:3456
URL of the COMU Agent Runtime server
comu.defaultModel
nvidia-nemotron-3-ultra
Default model ID used for agent sessions
🛡️ Security & Privacy
No Key Reselling: COMU does not provide or resell inference credits. You connect directly to your provider accounts.
Local Secret Storage: Keys are stored in VS Code's encrypted keychain (SecretStorage).
Workspace Concurrency: File edits utilize Optimistic Concurrency Control (OCC) and diff engines to protect your local workspace from corruption.
