esengine/DeepSeek-Reasonix

What it solves

Reasonix is a terminal-based AI coding agent designed to reduce token costs and friction in long coding sessions. It provides a unified engine that can power a CLI/TUI, a desktop application, and a VS Code extension, allowing developers to automate coding tasks directly from their workspace.

How it works

The project is built as a single static Go binary that uses a configuration-driven approach via reasonix.toml to manage providers, models, and plugins. It is specifically tuned for DeepSeek's prefix cache to keep costs low. The agent can use a combination of models (e.g., a planner and an executor) and extends its capabilities through plugins that run as subprocesses using MCP-compatible JSON-RPC over stdio.

Who it’s for

It is designed for developers who want a high-performance, low-cost AI coding assistant that integrates into their terminal, desktop, or IDE (VS Code).

Highlights

• Cache-aware context maintenance: Prunes stale tool output and injects environment summaries to optimize token usage.
• Config-driven architecture: Providers and tools are declared in TOML, avoiding hardcoded models.
• Multi-interface support: Available as a CLI/TUI, a standalone desktop app, and a VS Code extension.
• Plugin-driven extensibility: Supports external tools via MCP-compatible JSON-RPC.
• Zero-friction distribution: Distributed as a single binary with no dependencies other than a TOML parser.