Agnt is a minimalist, autonomous AI agent CLI tool written in the Odin programming language. It follows a modular, hexagonal architecture designed for extreme extensibility, privacy, and efficiency.

• Hexagonal Architecture: Ports and Adapters for LLM providers and communication channels.
• Unified Provider: Connects to ANY OpenAI-compatible API (OpenRouter, Groq, LM Studio, Ollama, Z-AI).
• Multi-Channel: Interact via terminal (CLI) or Telegram Bot.
• Interactive Setup: Simple onboarding with --setup command and model discovery.
• Hybrid Memory System:
  • Static: Loads context from .md files in your workspace (IDENTITY.md, SOUL.md).
  • Learned: Persistent facts stored in a local SQLite database.
  • Graph (Barabási-style): Semantic relations between entities (subject-predicate-object) mapped in SQLite.
• Agent Tools (Hands): Autonomous file I/O and real-time web search/fetch capabilities.

• Odin Compiler
• libsqlite3-dev (sudo apt install libsqlite3-dev)
• curl

git clone https://github.com/manelsen/agnt
cd agnt
# Build normal
odin build src -out:agnt
# Build otimizado (apenas ~450KB)
odin build src -out:agnt -o:speed
strip agnt
./agnt --setup

Agnt can run as a background service on Linux:

• agnt service install: Installs the systemd user service.
• agnt service uninstall: Removes the service.
• agnt service start / stop / restart: Controls the service.
• agnt service status: Checks if the agent is running.

Agnt keeps your system clean by isolating its brain in:

• Config: ~/.agnt/config/.env
• Memory: ~/.agnt/workspace/

• Multi-Agent Protocol: Allow Agnt to delegate tasks to other sub-agents and coordinate complex workflows.
• Shell & Python Tools: Autonomous shell command execution and a sandboxed Python interpreter for data processing.
• Semantic Vector Memory: Integration with sqlite-vec for RAG.
• Voice Interface: Audio-to-text and text-to-speech adapters.
• Graph Visualization: Export mental maps to Obsidian/Logseq.

MIT