Reasonix is a DeepSeek-native AI coding agent for the terminal, engineered around prefix-cache stability so you can leave it running. It is a config- and plugin-driven harness shipped as a single static Go binary — no Node, no Python, no runtime to install — tuned so that long sessions stay cheap on DeepSeek's token pricing.
Key Features
- Single Go Binary: CGO-free static binary, cross-compiled for darwin/linux/windows × amd64/arm64. The only dependency is a TOML parser.
- Prefix-Cache Tuned: Stable environment summary at startup, stale tool output pruned before compaction, and append-only prompt structure keep the KV-cache hit rate high — long autonomous runs stay affordable.
- Config-Driven: Providers, agents, tools, and plugins are all declared in
reasonix.toml. No hardcoded models; any OpenAI-compatible endpoint is a config entry. - Multi-Model & Composable: DeepSeek ships as a preset; optionally run two models together (executor + planner) in separate, cache-stable sessions.
- Plugin-Driven: MCP servers contribute tools, prompts, and resources; Extension Protocol v1 sidecars can intercept runtime events and ship versioned plugin packages.
- Full Tool Suite: readfile, writefile, editfile, movefile, bash, ls, glob, grep — with per-call permission policies (allow/ask/deny).
- Safety & Control: Plan mode (read-only), permissions, workspace sandbox, and per-turn checkpoints you can revert at any time.
- Four Entry Points: Terminal, desktop app, browser, or your editor over ACP (Agent Client Protocol). SSH remote sessions included.
Use Cases
Who Should Use This Tool?
- DeepSeek Power Users: Developers who want a coding agent tuned for DeepSeek's pricing model and 1M-token context.
- Terminal Purists: Anyone who wants a fast, zero-runtime coding agent without Electron bloat.
- Long-Running Automation: Teams that leave agents working on multi-hour tasks and need checkpoints and undo.
Problems It Solves
- Token cost on long sessions: Prefix-cache-aware context maintenance keeps repeat context cheap on DeepSeek.
- Runtime dependency hell: One static binary, six targets, nothing to install underneath.
- Unrecoverable agent runs: Per-turn checkpoints let you revert any turn and audit what the agent did.
Pricing
| Plan | Price | Features |
|---|---|---|
| Open Source | $0 | MIT-licensed; bring your own DeepSeek API key; pay only model usage. |
Advantages & Unique Selling Points
Compared to Competitors:
- DeepSeek-native economics: Purpose-built around prefix caching rather than treating DeepSeek as a generic OpenAI-compatible backend.
- Zero-friction distribution: A single static binary versus Node-based agents that need a runtime.
- Truly config-driven: No hardcoded models; every provider and tool is a TOML entry.
What Makes It Stand Out:
- 32,000+ GitHub stars and 3,000+ merged PRs with 98 contributors, developed in the open.
- Cache-stable multi-model collaboration (executor + planner).
- Plan mode, permissions, sandbox, and checkpoints make long autonomous runs readable and reversible.
Getting Started
Quick Start Guide
- Build:
git clone https://github.com/esengine/DeepSeek-Reasonix && cd DeepSeek-Reasonix && make build→bin/reasonix. - Configure: Set
DEEPSEEK_API_KEY; adjustreasonix.tomlfor providers and tools. - Run:
reasonixin your project directory (orreasonix -p "prompt"headless).
Integration
Integrates with:
- DeepSeek API (and any OpenAI-compatible endpoint)
- MCP servers for tools, prompts, and resources
- Zed and other editors via ACP
Frequently Asked Questions
Does it work with models other than DeepSeek?
Yes. DeepSeek ships as a preset, but any OpenAI-compatible endpoint is a config entry.
How does it keep costs low?
It maintains a stable prompt prefix and prunes stale tool output before compaction, maximizing DeepSeek's prefix-cache (KV-cache) hit rate.
Is it really a single binary?
Yes — CGO-free Go, cross-compiled for darwin/linux/windows × amd64/arm64. Nothing else needs installing.
Alternatives
If Reasonix isn't the right fit, consider these alternatives:
- Todos: Chief-agent orchestration workspace supporting DeepSeek and 700+ models.
- Grok Build: xAI's terminal coding agent powered by grok-4.5.
- CrewAI: Python role-based multi-agent framework.
Tips & Best Practices
- Keep sessions running: Reasonix is built to be left running; the prefix cache rewards long-lived sessions.
- Use plan mode for exploration: Read-only investigation keeps costs predictable before edits begin.
- Checkpoint before risky turns: Revert any turn with per-turn checkpoints if something goes wrong.
Conclusion
Reasonix is the terminal coding agent that treats DeepSeek as a first-class citizen — prefix-cache-tuned, config-driven, and shipped as one static Go binary with 32,000+ stars. If you live in the terminal and work with DeepSeek, it's one of the most cost-effective agents you can run.
Comments
No comments yet. Be the first to comment!
Related Tools
FastClaw
fastclaw.ai
Lightweight AI Agent runtime and Multi-Agent Framework written in Go. An Agent Factory that creates, manages, and runs AI agents with personalities (SOUL.md), memory, skills, and tools — single binary, any LLM, cloud-ready.
Grok Build
docs.x.ai/build/overview
xAI's terminal-based AI coding agent powered by grok-4.5. A full-screen TUI that understands your codebase, edits files, runs shell commands, searches the web, and manages long-running tasks — interactively, headlessly, or via ACP.
Todos
todos.dev
Task-driven workspace where a Chief agent splits your goal into todos, assigns each to agents by trade, and tracks progress — supporting 700+ models including DeepSeek, with execution on your own machines.
Related Insights
After I Connected Obsidian to OpenClaw, It Started Helping Me Make Decisions
Once Obsidian stopped being just a place to store notes and started working with OpenClaw, it began helping me organize context, connect information, and improve real decisions.
Stop Cramming AI Assistants into Chat Boxes: Clawdbot Picked the Wrong Battlefield
Clawdbot is convenient, but putting it inside Slack or Discord was the wrong design choice from day one. Chat tools are not for operating tasks, and AI isn't for chatting.
The Twilight of Low-Code Platforms: Why Claude Agent SDK Will Make Dify History
A deep dive from first principles of large language models on why Claude Agent SDK will replace Dify. Exploring why describing processes in natural language is more aligned with human primitive behavior patterns, and why this is the inevitable choice in the AI era.