Local-first AI developer tools for code intelligence, MCP integrations, efficient LLM inference, and model compression.

VeyrForge builds practical tools that help developers understand large codebases, run AI models locally, and reduce the memory required for local inference.

A local-first MCP server and repository intelligence engine for Cursor, Claude Code, Codex, and other AI coding assistants.

Codehelper indexes source code into symbol and call graphs, providing code search, impact analysis, dependency tracing, test-impact detection, and more than 60 MCP tools without uploading the entire repository to a cloud service.

Best for: AI coding agents, codebase understanding, repository indexing, static analysis, and local development workflows.

A memory-smart local LLM inference runtime for consumer CPUs and GPUs.

Green Engine provides MoE expert scheduling, KV-cache paging, long-context management, GGUF model support, and local OpenAI-compatible inference through the ge CLI.

Best for: Local LLM inference, low-VRAM systems, long-context workloads, coding agents, and offline AI.

A Rust toolkit for post-training LLM weight compression and low-memory inference.

Green Compress supports Q4/Q8 quantization, adaptive weight repair, GGUF model processing, AVX2 CPU inference, and optional CUDA acceleration.

Best for: Model compression, quantization, lower RAM usage, CPU inference, and preparing models for Green Engine.

Codehelper
    │
    ├── Repository indexing
    ├── Symbol and call graphs
    ├── MCP tools
    └── Code impact analysis
            │
            ▼
Green Engine
    │
    ├── Local LLM inference
    ├── Embeddings
    ├── Chat and routing
    └── Memory-aware scheduling
            │
            ▼
Green Compress
    ├── Weight compression
    ├── Q4/Q8 quantization
    └── Reduced model memory usage

The tools can be used independently or together as a completely local AI development stack.

New to VeyrForge?

1. Start with Codehelper for local repository intelligence and MCP tools.
2. Add Green Engine when you want local embeddings or local chat.
3. Use Green Compress when model memory usage becomes the bottleneck.

VeyrForge projects are publicly available as source-available software. Public development, issues, releases, and changelogs are maintained in the official VeyrForge repositories.

Bug reports, benchmark results, compatibility reports, and suggested improvements are welcome.