DeepSeek-Reasonix: DeepSeek‑Reasonix: a config‑driven, plugin‑based AI coding agent for the terminal.

DeepSeek-Reasonix is a terminal‑based AI coding agent that works with DeepSeek (or any OpenAI‑compatible) models. It is distributed as a single static Go binary, built with CGO_ENABLED=0, and tuned around DeepSeek’s prefix cache so token usage stays low across long sessions.

How it works

  • Configuration lives in reasonix.toml where providers, the agent, enabled tools, and plugins are declared.
  • External tools run as subprocesses communicating over stdio JSON‑RPC (MCP‑compatible); built‑in tools self‑register at compile time.
  • The agent keeps a cache‑aware context: at startup it injects a small stable environment summary, stale tool output is snipped/pruned before summary compaction, and the tool schema contract is documented for regression review.
  • It can run two models together (e.g., an executor and a planner) in separate, cache‑stable sessions.

Key features

  • Config‑driven – no hardcoded models.
  • Multi‑model & composable – DeepSeek preset plus any OpenAI‑compatible endpoint; optional dual‑model setup.
  • Plugin‑driven – external tools via stdio JSON‑RPC.
  • Cache‑aware context maintenance – stable summary, pruning, schema contract.
  • Zero‑friction distribution – single binary, cross‑compiles to six targets with one command.

Usage Install the binary (e.g., npm i -g reasonix or Homebrew, or use prebuilt archives). Then run:

reasonix setup          # configure provider and model
reasonix                # start interactive session
reasonix run "implement the TODOs in main.go"

In an interactive session, /init creates project instructions. A desktop app and a VS Code extension are also available.

Limitations The README does not mention any specific limitations.

Sources