v0.1.0

Wildcode (Beta)

Wildcode is an advanced, memory-frugal agentic CLI coding assistant designed specifically for local, privacy-first AI models.

Built on a Node-based orchestration pipeline, Wildcode completely rethinks how AI agents interact with massive codebases. Instead of relying on massive context windows that exhaust VRAM, it trades "Disk/Time for Memory" to provide a robust, pair-programming experience that runs fluently on memory-constrained devices (like Mac M-series running Gemma4-26B).

🔥 Core Highlights in this Release

• The Polymath Expert Tribunal: A multi-agent review loop where a Lead Engineer drafts code and a Technical Reviewer strictly audits it. If a draft fails the AI-Driven Polyglot Scoped Compilation, it is VETOED before touching your disk.
• Deep Semantic Indexing: Beyond simple RAG. Wildcode parses your AST, groups files into logical modules, and builds a token-budgeted Map-Reduce architecture to handle ultra-large files.
• Deterministic Patch Application: "Intelligence" is strictly separated from "Execution". AI generates patches in a strict DSL, ensuring changes are perfectly confined to the approved writeScope.
• Pure Ephemeral State: Zero footprint in your project repo. Sessions, drafts, and AST trees are purely in-memory and garbage-collected instantly to minimize RAM usage.

🛠️ Getting Started

Ensure you have pnpm installed, then:

# Install and build
pnpm install && pnpm build

# Start the interactive CLI
pnpm dev

(For detailed configuration and slash commands like /init and /model, please refer to the README.)

⚠️ Beta Disclaimer

This project is in active development. APIs, CLI commands, and internal architectures are highly unstable. Please ensure you have committed your work to version control before running Wildcode.