dirac-run/dirac

Coding Agent singularly focused efficiency and context curation. Reduces API costs by 50-80% vs other agent AND improves the code quality at the same time. Uses Hash Anchored edits, massively parallel operations, AST manipulation and many many other optimizations. https://dirac.run/

What it solves

Dirac addresses the inefficiencies of AI coding agents during complex, long-running software engineering tasks. It reduces token overhead, minimizes "approval fatigue" for the user, and prevents the model from drifting during multi-day objectives by providing a more robust infrastructure for autonomous execution.

How it works

Dirac operates as an agent that can be integrated into VS Code, the terminal, or any Agent Client Protocol (ACP) compatible editor. It uses several specialized mechanisms to improve precision and efficiency:

  • Hash-Anchored Editing: Uses a stable line-anchor protocol instead of search-and-replace to make file edits more precise.
  • AST Inspection & Manipulation: Uses syntax trees to outline files or rename symbols across a codebase without reading every line of text.
  • Utility Models: Routes low-complexity tasks (like commit messages or permission checks) to cheaper models to save costs.
  • Goal Mode: Allows users to set a high-level objective that the agent pursues autonomously, pausing only for necessary input.
  • Subagents: Deploys concurrent agents to research and validate work in parallel.
  • Self-Extension: Can create its own typed tools via a /new-tool command to adapt to specific workflows.

Who it’s for

It is designed for software engineers who need an AI agent capable of handling complex refactoring, long-term feature development, and large-scale codebase changes across various IDEs and terminal environments.

Highlights

  • Multi-Interface Support: Works in VS Code, Open VSX, CLI, and ACP clients (like Zed and JetBrains).
  • Model Agnostic: Supports dozens of providers and hundreds of models.
  • Parallel Operations: Can batch multiple reads, searches, and edits in a single model response.
  • Context Management: Features context condensation and restore points to maintain continuity in long tasks.
  • Automated Permissions: Uses a utility model to automatically approve requests based on a natural-language policy.

Related

  • Project
    syrizelink/OpenFic

    An AI-native writing tool for long-form fiction that uses Agentic RAG and customizable workflows to help authors maintain world-building and character consistency.

  • Project
    clacky-ai/openclacky

    An open-source, token-efficient AI agent that reduces costs through high cache hit rates and a minimal toolset, supporting any OpenAI-compatible model for coding and general tasks.

  • Project
    zdk/lowfat

    A lightweight CLI tool that reduces AI token costs by filtering CLI output and compressing file content before it reaches an AI agent.

  • Project
    tontinton/maki

    An AI coding agent optimized for minimal context token usage and high performance, featuring a fast TUI and support for numerous LLM providers.

  • Project
    Houseofmvps/codesight

    A zero-dependency tool that pre-compiles codebase structure and documentation into compact context maps and wikis to drastically reduce token consumption for AI coding assistants.