Overview
A budget-conscious, model-agnostic AI agent CLI that talks to models via OpenRouter. Features autonomous multi-step execution, Model Context Protocol (MCP) stdio client/server, local SQLite session persistence with real-time cost tracking, multi-agent consensus debate, and full-screen Textual TUI dashboard.
Key Features
- ✓Autonomous Plan → Act → Reflect execution loop with LLM reflection
- ✓Model Context Protocol (MCP) bidirectional JSON-RPC integration
- ✓Modular sub-agents with Git branch and worktree isolation
- ✓Multi-agent consensus debate and recursive peer delegation
- ✓Dynamic token budget ceilings and real-time session spend tracking
- ✓Full-screen interactive TUI dashboard with telemetry gauges
- ✓Interactive slash commands (/budget, /model, /goal, /tokens, /cost)
Outcomes & Impact
- →Zero vendor lock-in with support for 200+ open-source and proprietary models
- →Deterministic sub-cent cost enforcement preventing runaway LLM billing
- →Complete task autonomy without context window bloat via LRU cache
Tech Stack
Python 3.10+
backend
MCP Protocol
other
OpenRouter LLMs
other
SQLite
database
Textual / Rich TUI
frontend
Git Worktrees
devops
Tags
Python
MCP
OpenRouter
SQLite
Textual TUI
LLM Agents
Git Worktrees
Project Info
Categories
AI/ML, backend, open-source
Status
Public