Skip to content

πŸ’ Lychee v0.1.0-alpha β€” First Public Release

Pre-release
Pre-release

Choose a tag to compare

View all tags
@MD-Mushfiqur123 MD-Mushfiqur123 released this 11 Jun 12:05
· 78 commits to main since this release
v0.1.0-alpha
865198c

πŸ’ Lychee v0.1.0-alpha

What is Lychee?

Lychee is a drop-in upgrade for Ollama β€” a local LLM runtime with enterprise-grade features built on top of the battle-tested Ollama inference engine.

✨ What's New in This Release

πŸ”Œ Multi-Protocol API Gateway

  • Anthropic Messages API (/v1/messages) β€” run Claude-compatible workloads locally with any model
  • OpenAI Responses API (/v1/responses) β€” structured JSON output with full JSON Schema validation
  • OpenAI Chat Completions (/v1/chat/completions) β€” drop-in OpenAI replacement

πŸ€– Model Composer (/api/compose)

Chain multiple local models together in a pipeline:

  • Sequential execution with template substitution ({{input}}, {{step[n].output}})
  • Parallel branch execution β€” run independent steps concurrently
  • Per-step timeouts β€” prevent hanging pipelines
  • Fallback model β€” automatically retry with a backup model on failure
  • SSE streaming β€” real-time token streaming for each step

πŸ›‘οΈ Stateful Agent Sandbox

  • Terminal UI approval gates for every tool call
  • Auto-allow safe commands, deny dangerous patterns (fork bombs, privilege escalation)
  • Path traversal prevention, hierarchical allowlisting

🌐 Embedded Web Dashboard (/dashboard/)

  • Full React SPA served directly from the binary
  • Chat interface, model management, system resource monitoring

πŸ“₯ Direct HuggingFace Pull (\lychee hf pull)

  • Download GGUF models directly from HuggingFace Hub
  • Multi-shard concurrent downloads with resume support
  • SHA256 integrity verification

πŸ“Š Developer Tools

  • \lychee compare\ β€” benchmark multiple models side-by-side
  • \lychee stats --tui\ β€” live terminal performance dashboard
  • \lychee scan\ β€” hardware detection and model recommendations
  • \lychee generate-client\ β€” generate SDK boilerplate in Python/JS/Go/Rust

πŸš€ Quick Start

\\οΏ½ash
git clone https://github.com/MD-Mushfiqur123/lychee.git
cd lychee
go build -o lychee .
./lychee serve
\\

πŸ“‹ Requirements

  • Go 1.22+
  • A GPU (NVIDIA/AMD/Apple Silicon) or CPU-only mode

⚠️ Alpha Notice

This is an alpha release. Core features are functional but not yet production-hardened. Use with care in production environments.

Built on top of Ollama β€” the open-source local LLM runner.

Assets 2
Loading