esengine/DeepSeek-Reasonix

What it solves

Reasonix is an autonomous coding agent designed for long-running tasks. It addresses the difficulty of managing autonomous AI runs by providing mechanisms to track, review, and undo changes, ensuring that a developer can leave the agent running without losing control over the codebase.

How it works

Reasonix operates as a local engine written in Go, distributed as a single static binary. It can be accessed via a terminal (CLI/TUI), a desktop application, a web browser, or directly within an editor like VS Code via the ACP (Agent Communication Protocol).

Key technical components include:

• Config-driven Architecture: Providers and models are defined in a reasonix.toml file, allowing for easy switching between OpenAI-compatible endpoints.
• Composable Models: It supports running separate models for planning and execution to maintain cache stability.
• Plugin System: It integrates with MCP servers for tools and prompts, and uses an Extension Protocol for runtime event interception and UI contributions.
• Context Management: It uses a cache-aware system that prunes stale tool output and injects environment summaries to maintain efficient context.
• Safety Features: It includes a workspace sandbox, per-turn checkpoints for rewinding, and a permission system.

Who it’s for

It is built for developers who want an autonomous AI coding assistant that can handle complex, multi-step implementation tasks while remaining transparent and manageable.

Highlights

• Multi-interface access: Use it via CLI, desktop app, or VS Code extension.
• Zero-friction setup: Distributed as a single Go binary with no external dependencies on the target machine.
• Flexible model support: Works with any OpenAI-compatible endpoint and supports planner/executor splits.
• Extensible: Supports MCP servers and a dedicated Extension Protocol for custom plugins.
• Safe autonomy: Features a workspace sandbox and the ability to undo changes via checkpoints.