1ay1/agentty

AI pair programming in your terminal — one static binary, sub-ms startup, any model

What it solves

agentty is a high-performance terminal-based coding agent designed to replace slower, runtime-dependent tools (like those built on Node.js or Python). It solves the problem of "context dumping," where agents send entire repositories to an LLM, by using a local retrieval system to send only the most relevant code slices, reducing context usage and costs.

How it works

Written in C++26 as a single static binary, agentty operates as a native application with millisecond startup times. It employs a sophisticated local RAG (Retrieval-Augmented Generation) engine that combines BM25 keyword search with dense embeddings and GraphRAG to fetch precise code and documentation. It supports a wide array of LLM providers (Claude, OpenAI, Groq, Ollama, etc.) and features a "Smart Mode" that uses an orchestrator-worker architecture to delegate tasks between flagship and cheaper models based on complexity.

Who it’s for

It is built for developers who want a fast, vendor-agnostic coding assistant that runs directly in the terminal, works offline (via local models), and maintains strict security through sandboxed shell execution.

Highlights

  • Native Performance: Single static binary with ~3ms cold start and zero runtime dependencies.
  • Advanced Local RAG: Hybrid retrieval (BM25 + embeddings), GraphRAG, and reranking to minimize context window bloat.
  • Model Agnostic: Supports multiple providers including Claude, OpenAI, DeepSeek, and local Ollama models with live switching.
  • Security First: Default sandboxing for shell/build commands and an air-gapped mode for secure remote execution via SSH.
  • Smart Mode: An orchestrator-worker system that scales effort and model cost based on the complexity of the task.
  • Zed Integration: Compatible with the Agent Client Protocol (ACP) for use inside the Zed editor.

Related

  • Project
    Dicklesworthstone/pi_agent_rust

    A native Rust-based AI coding agent CLI that provides a high-performance, secure, and resource-efficient alternative to managed-runtime AI assistants.

  • Project
    AgentEra/Agently

    An AI application runtime framework that provides structured output guarantees, observable action execution, and signal-driven workflows for building reliable AI service backends.

  • Project
    Prof-Harita/terminaI

    A local-first AI terminal operator that converts natural language intent into governed, reviewable, and reversible system actions with built-in safety guardrails.

  • Project
    shibing624/agentica

    Agentica is an open‑source, Apache‑2.0‑licensed Python framework that lets you run LLM‑based agents locally. A single chat session becomes an autonomous agent that can call built‑in tools (search, file I/O, code execution, OCR, image generation), remember past work, and spawn sub‑agents or delegate whole tasks. It works via a CLI, a local web UI, a desktop app, and IM gateways (WeChat, Feishu, Telegram). Supports many model providers (OpenAI, DeepSeek, Claude, Ollama, etc.), offers async‑first APIs, safety guardrails, multimodal input, persistent memory, a markdown‑based skill system, and RAG integration. Install with `pip install -U agentica` (add `[gateway]` for the web UI) and configure an API key. Ideal for developers, researchers, and power users who want a self‑hosted “LLM OS” with collaborative multi‑agent capabilities.

  • Project
    jnMetaCode/agency-orchestrator

    Agency Orchestrator is a Node‑JS tool that lets you run dozens of AI “expert” roles together. You can write a short YAML workflow or simply give a one‑sentence prompt; AO picks the right roles, builds a DAG, calls LLM providers (Claude, DeepSeek, OpenAI, etc.), runs steps in parallel, and produces a final report. It includes a web Studio UI, role libraries (267 Chinese roles + multilingual packs), support for key‑less providers, acceptance criteria, resumable runs, and integrations with coding assistants via an MCP server.