Deepseek Reasonix
DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.
Last verified:
What is Deepseek Reasonix?
Deepseek Reasonix is an open-source AI coding agent designed specifically for your terminal, engineered around DeepSeek's prefix-cache technology to keep token costs low across long coding sessions. It communicates directly with api.deepseek.com and uses an append-only history loop that preserves the byte prefix across every tool call, enabling cache hit rates over 90% and reducing input token costs to around one-fifth of non-cached usage.
Deepseek Reasonix pricing
Pricing model: Freemium
Reasonix itself is free and open-source under MIT license with no paid plans. However, it requires a DeepSeek API key from platform.deepseek.com and you pay DeepSeek directly for API tokens used. The cache-first architecture significantly reduces costs by achieving 90%+ cache hit rates, reducing input token costs to approximately one-fifth of what you would pay without caching. There is no free tier from DeepSeek mentioned - you pay per token usage through DeepSeek's pricing.
Deepseek Reasonix pros
- Open-source MIT license with community development
- Cache-first loop achieves 90%+ cache hit rates
- Input token costs reduced to ~1/5 of non-cache usage
- Terminal-first design - leave it running for long sessions
- Append-only history preserves byte prefix across tool calls
- Custom cell-diff renderer with no Ink dependency
- Plan mode lets you review edits before applying to disk
- Granular per-tool permissions (allow/ask/deny)
- Embedded dashboard shows live cache hit rate and costs
- Persistent resumable sessions per workspace
- First-class MCP support with stdio and HTTP transports
- Auto-checkpoints provide cursor-style rollback for AI edits
- /effort knob adjusts reasoning depth per turn
- Semantic search with local Ollama or DeepSeek embeddings
- Transcript replay useful for bug reports and demos
- Event log enables custom dashboards and analytics
- No install needed - works with single npx command
- R1 thought harvesting distills reasoning into typed plan state
- Tool-call repair handles DeepSeek-specific quirks automatically
Deepseek Reasonix cons
- DeepSeek-only by design - cannot use other AI models
- Requires Node ≥ 22 which may be newer than some systems have
- Needs DeepSeek API key - not free to use (pay per token)
- Windows requires PowerShell, Git Bash, or Windows Terminal
- First-time setup requires wizard for API key and preset selection
- No graphical UI - terminal-only interface
- Community-built means fewer official support resources
- MCP server configuration requires learning new spec format
Frequently asked questions about Deepseek Reasonix
What is Reasonix?
Reasonix is a DeepSeek-native AI coding agent for your terminal. It is engineered around DeepSeek's prefix-cache so token costs stay low across long sessions. It uses an append-only history loop that preserves the byte prefix across every tool call, enabling DeepSeek's prefix-cache to keep hitting turn after turn.
How do I get started with Reasonix?
First, get a DeepSeek API key from platform.deepseek.com. Then navigate to your project directory and run 'npx reasonix code'. No install is needed. The first run walks you through a short wizard where you paste your API key, pick a preset, and optionally attach MCP servers. The agent then proposes edits as reviewable blocks.
What operating systems does Reasonix support?
Reasonix requires Node ≥ 22 and runs on macOS, Linux, and Windows. On Windows, you need PowerShell, Git Bash, or Windows Terminal. You can press Esc anytime to abort and type '/help' for the full slash-command list.
How does the cache-first loop work?
The cache-first loop uses append-only history with no in-place mutation or marker-based compaction. The byte prefix survives every tool call, so DeepSeek's prefix-cache keeps hitting turn after turn. This is different from generic agent frameworks that were designed for different cache mechanics.
What is plan mode?
Plan mode lets you review proposed edits before they touch disk. You can approve, refine, or reject changes. Plan checkpoints persist across runs so you can resume mid-review. Nothing hits disk until you type '/apply'.
How do I control costs with Reasonix?
Reasonix manages costs through cache-safe folding, aggressive-fold tier, summary-on-exit, and model-aware budgets. The loop manages context size without breaking prefix stability. Cache hit rates over 90% reduce input token costs to around one-fifth of non-cache usage.
What is the /effort knob?
The /effort knob lets you switch reasoning depth per turn. Use 'max' for difficult problems and 'low' for routine tasks. It is available as both a slash command and a CLI flag '--effort <max|high|medium|low>'.
Are sessions persistent in Reasonix?
Yes, sessions are per-workspace, named, and resumable. Using '--resume' picks up exactly where you left off, including system prompt, history, and plan state. This lets you leave Reasonix running for long coding sessions.
Is Reasonix open source?
Yes, Reasonix is MIT licensed and community-developed. It has scoped 'good first issue' tickets with code pointers and acceptance criteria, and real PRs from real contributors. Every contributor avatar on the website represents a real PR that shipped.