DeepSeek Reasonix: Native Coding Agent with High Caching and Low Cost
Reasonix is a new terminal-based AI coding agent built natively for DeepSeek models. It's designed around two key constraints: high caching efficiency and low cost — both critical for developers running code generation loops frequently.
Key Details
- Native DeepSeek support — Reasonix is tuned specifically for DeepSeek's architecture, not wrapped around a generic API. This allows optimizations like KV-cache reuse across consecutive requests.
- Aggressive caching — The tool caches intermediate results (prompt embeddings, partial completions) to avoid redundant computation. Early reviews mention “near-instant warm restarts” after the first call.
- Cost — Paired with DeepSeek's newly permanent V4 Pro price discount (from $0.50/M tokens → $0.17/M tokens, per HN thread), Reasonix claims “best-in-class per-request cost” for an agentic coding loop.
- Terminal-native — Runs as a CLI tool (
reasonix --model deepseek-v4-pro). No IDE plugin required. Supports both streaming and batch mode. - Open source — Repository at esengine.github.io/DeepSeek-Reasonix.
📖 Read the full source: HN AI Agents
👀 See Also
Architect CLI: Open-source tool for orchestrating headless AI coding agents in CI/CD
Architect is an open-source CLI tool designed for autonomous AI coding agents in CI/CD pipelines, featuring the Ralph Loop for test-retry cycles, deterministic guardrails, YAML pipeline definitions, and support for multiple LLMs via LiteLLM.
PreToolUse Hook Fixes Claude Code Image Crash Problem
A developer created a PreToolUse hook that intercepts Claude Code's Read calls on images, converts them safely, and proxies them through a Haiku subprocess to prevent API Error 400 crashes from problematic images.
Vibeyard adds P2P session sharing for Claude Code
Vibeyard, an open-source IDE for Claude Code, now supports peer-to-peer session sharing. Users can share live terminal sessions with teammates over encrypted WebRTC connections with read-only or read-write access modes.
Engram: Open-source memory layer for Claude Code and MCP clients
Engram is an open-source memory layer that works as an MCP server with any client like Claude Code, Cursor, or Windsurf. It stores unlimited memories with semantic vector search, achieves 80% accuracy on LOCOMO benchmark, and uses about 800 tokens per query versus 5K+ for file-based approaches.