dirge-code/dirge

Dynamic Intent Resolution Grounding Engine

What it solves

Dirge is a high-performance, minimal coding agent designed to provide long-term continuity across sessions while keeping the operational cost and resource usage low. It addresses the common issue of agents "forgetting" context between sessions and the high memory overhead of JavaScript-based agentic editors.

How it works

Built in native Rust, Dirge uses a two-tiered memory system to maintain project-specific and global user preferences without relying on vector embeddings. Instead, it utilizes SQLite FTS5 for session search and plain grep for code search, arguing that lexical search is more reliable for weaker models. It manages a bounded context budget (default 250k tokens) by writing resumable checkpoints and distilling salient facts into a durable session database.

Who it’s for

Developers who want a fast, lightweight coding agent that can work with a variety of LLM providers (including local models via Ollama) and requires robust permissioning and long-horizon session management.

Highlights

  • Extreme Efficiency: Uses significantly less RAM (approx. 8-15 MB) compared to JS-based agents.
  • Robust Agent Loop: Includes a system to repair malformed tool calls and validate writes via tree-sitter before they touch the disk.
  • Multi-Provider Routing: Allows different models to be assigned to specific roles like review, escalation, and summarization within a single session.
  • Integrated Issue Board: Features a built-in agent-facing kanban board to manage backlogs without external trackers.
  • Extensible Architecture: Supports a Janet plugin system and Claude-compatible skills for runtime extensibility.
  • MCP Server Support: Can be exposed as an MCP server, allowing other agents (like Claude Code) to delegate implementation tasks to it.

Related

  • Project
    gi-dellav/zerostack

    zerostack is a lightweight coding agent written in Rust that runs in the terminal, supporting multiple LLM providers. It offers granular permission controls, session management, subagents, and optional persistent memory, while using far less RAM and CPU than JavaScript-based agents.

  • Project
    sopaco/cortex-mem

    A Rust-based long-term memory framework for AI agents that uses a three-tier hierarchical architecture and a virtual filesystem to provide persistent, context-aware interactions.

  • Project
    pacifio/cersei

    A complete Rust SDK for building high-performance coding agents with built-in tool execution, graph memory, and sub-agent orchestration.

  • Project
    deepagent-ltd/deepagent-code

    DeepAgent Code (DAC) is a desktop‑plus‑terminal workspace that lets you use an LLM as a coding assistant for multi‑step tasks. It offers persistent sessions, inspectable project memory, sub‑agents for independent jobs, a managed browser, and optional instant‑messaging integration, all configurable with any compatible LLM provider. Licensed under AGPL‑3.0.

  • Project
    vinhnx/VTCode

    VTCode is a terminal-based open-source AI coding agent (Rust) that plans, builds, and reviews code with hosted or local models — featuring an interactive TUI, headless exec mode, sandboxed tool execution, resumable sessions, and a self-correcting "HOVER" loop.