Codmir Assistant Vision

Codmir is evolving from an AI coding companion into a local-first assistant that understands context, listens for your voice, sees the screen, and safely controls the desktop—while keeping data on your machine.

Principles

• Local-first: run core components offline (wake word, transcription, TTS, capture)
• Privacy-by-design: visible indicators (blue corners), explicit arming windows, and logs
• Composable: small workers, simple APIs, dockerized
• Interoperable: can route through n8n or direct HTTP
• Safe control: dry-run modes, per-action confirmations, and time-bound sessions

Capabilities

• Wake word “codmir” to start a session
• Low-FPS, change-aware screen capture with overlay indicator
• Microphone recording and offline speech-to-text (Vosk)
• Desktop actions: open URLs/apps, type, keys, clicks, focus windows
• Session persistence: audio + frames + transcript + metadata for review

Presentation Mode

• Demonstrate abilities on command (e.g., “open our app” on http://localhost:3000)
• Optional TTS responses (local espeak-ng by default; cloud providers documented only)

Safety

• Overlay shows when Codmir “can see the screen”
• Arm/disarm phrase for control (time-limited)
• Dry-run mode to preview actions without changing state
• Clear audit trail in session artifacts

Roadmap

• Wayland-native desktop control backends
• Rich intent routing and confirmations
• Pre/post-roll audio buffers and frame timelines
• In-app session viewer with search and summaries

Learn More

• Overview: Codmir Local Assistant
• Use Case: Voice Desktop Assistant (Local)
• Orchestration: Orchestrator