Implementation

Tech Stack

Project Structure

omar/
├── Cargo.toml              # Workspace root
├── src/
│   ├── main.rs             # Entry point, CLI, tmux relaunching
│   ├── app.rs              # Core state machine, refresh/render
│   ├── config.rs           # Configuration loading + auto-detection
│   ├── event.rs            # Input/tick event handling
│   ├── memory.rs           # Persistent state management
│   ├── computer.rs         # X11 computer use integration
│   ├── projects.rs         # Project CRUD
│   ├── api/
│   │   └── handlers.rs     # HTTP API endpoints (axum)
│   ├── tmux/
│   │   ├── client.rs       # tmux command wrapper
│   │   ├── session.rs      # Session types
│   │   └── health.rs       # Health checking
│   ├── manager/
│   │   └── mod.rs          # EA session lifecycle
│   ├── scheduler/
│   │   └── mod.rs          # Event scheduling + delivery
│   └── ui/
│       └── dashboard.rs    # TUI rendering
├── omar-slack-bridge/      # Slack bridge crate
├── omar-computer-bridge/   # Computer use bridge crate
└── prompts/                # Embedded prompt templates

Key Modules

main.rs (~743 lines)

Entry point with CLI parsing (clap). Auto-relaunches inside tmux if not already in a tmux session. Manages daemon lifecycle, bridge auto-spawning, and graceful shutdown (SIGTERM then SIGKILL).

app.rs (~1,298 lines)

Core application state machine. Manages agent list, selection, focus, projects, and UI state. Handles refresh cycles that poll tmux for session updates. Builds the hierarchical command tree for agent visualization.

api/handlers.rs (~859 lines)

Full REST API on axum with CORS. Endpoints for agent CRUD, event scheduling, computer use, and project management. Supports session name normalization (short names like “auth” resolve to “omar-agent-auth”).

scheduler/mod.rs

Event scheduling system using a BinaryHeap (min-heap by timestamp). Supports one-shot and recurring events. Delivers events by injecting text into agent tmux sessions via send-keys. Thread-safe with Arc<Mutex<>>.

memory.rs

Persistent state snapshots. Writes active state (projects, agents, tasks) to ~/.omar/memory.md. This gets injected into the EA prompt on startup for context continuity.

CLI Interface

# Start dashboard (default - auto-detects backend)
omar

# Use specific agent backend
omar --agent opencode

# Custom config
omar --config ~/.config/omar/config.toml

# Spawn a new agent
omar spawn --name my-agent --command "claude"

# List agents (non-TUI)
omar list

# Kill an agent
omar kill agent-1

# Start executive assistant
omar manager start

# Setup tmux configuration
omar setup-tmux

Auto-Detection

On startup, OMAR auto-detects available agent backends:

1. Checks for claude (v1+) in PATH
2. Falls back to opencode if available
3. Defaults to claude --dangerously-skip-permissions
4. Override with --agent <name>

Bridges

Both bridges are optional and auto-spawned when their environment is available:

• Slack Bridge: Spawns if SLACK_BOT_TOKEN and SLACK_APP_TOKEN are set. Routes Slack messages to OMAR agents via the API.
• Computer Bridge: Spawns if DISPLAY is set. Provides X11 mouse/keyboard/screenshot control with exclusive locking per agent.

Embedded Prompts

Prompts in /prompts/ are compiled into the binary and synced to ~/.omar/prompts/ on first run:

• executive-assistant.md - EA system prompt (gets memory context prepended)
• agent.md - Unified agent prompt (injected for all spawned agents)
• prompts/tests/ - Test scenarios like project-factory.md