Verified project record

esengine/DeepSeek-Reasonix

Reasonix is a terminal-based AI coding agent built in Go that is specifically tuned to reduce DeepSeek API token costs during long coding sessions through prefix-cache stability. It is distributed as a single static binary and uses an external DeepSeek API for inference, which means data is sent to an external service and usage may incur paid service costs.

Project overview

The harness is specifically engineered around DeepSeek's prefix cache to keep token costs low across long coding sessions, while offering a single-binary distribution model with plugin-driven external tools.

Project type: MCP · AI Agent · AI Coding
Use cases: Coding & Development · Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers and AI engineers who work from the terminal and need a configurable, composable agent harness tuned for DeepSeek models.

Key capabilities

Providers, the agent configuration, enabled tools, and plugins are declared in a reasonix.toml file with no hardcoded models.
DeepSeek ships as a default preset, but any OpenAI-compatible endpoint can be added as a config entry. The system can optionally run two models simultaneously (executor and planner) in separate, cache-stable sessions.
External tools run as subprocesses over stdio JSON-RPC using MCP-compatible protocols. Built-in tools self-register at compile time.
Startup injects a small stable environment summary, stale tool output is snipped or pruned before summary compaction, and the built-in tool schema contract is documented for regression review.
The harness compiles with CGO_ENABLED=0 as a single static Go binary and can cross-compile to six targets with one command. The only dependency is a TOML parser.
Supports slash commands and @ references for terminal interaction.
Users can create, share, preview, run, edit, and safely delete isolated agent profiles from the desktop application or command line.
Provides a snapshot-based edit safety net accessible via Esc-Esc or /rewind commands.

Limitations and risks

The tool is engineered specifically around DeepSeek's prefix cache stability, though other OpenAI-compatible models can be configured.
Requires external API access (such as DeepSeek), which may incur paid service costs.

Getting started

Install the single binary via npm or brew, set up a minimal reasonix.toml file, configure the DeepSeek API key via setup or environment variables, and run the reasonix command.

Alternatives and comparisons

A native, open source AI agent accessible via desktop app, CLI, or API that works with any LLM and connects to numerous extensions, supporting broader use cases beyond terminal coding.
An opinionated CLI providing out-of-the-box multi-agent orchestration with automated model routing, context injection, and specialized discipline agents for parallel task execution.
A self-hosted AI chat platform with multi-user support, agents, tools, and MCP support, suitable for users needing a conversational interface across multiple providers without vendor lock-in.

Project comparisons

Evidence and sources

GitHub project description: DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.
README: A config- and plugin-driven harness — a single static Go binary, tuned around DeepSeek's prefix cache so token costs stay low across long sessions.
README: A DeepSeek-native AI coding agent for your terminal.
README: Zero-friction distribution. CGO_ENABLED=0 single binary; cross-compile to six targets with one command. The only dependency is a TOML parser.
README: Windows builds are code-signed with a free certificate provided by the SignPath Foundation, with signing through SignPath.io.

Reviewed at: 2026-07-17T10:14:35Z

Open original GitHub project