Summary

Nyx addresses the need for a sophisticated, local-first AI coding agent interface within VS Code and Cursor. It targets developers seeking private, zero-token-cost AI assistance by seamlessly integrating powerful open-weight models running on user hardware, offering a Cursor-grade user experience without cloud dependencies.

How It Works

Nyx functions as an editor extension that prioritizes local inference, managing a fleet of user-hosted AI model endpoints (Ollama, LM Studio, etc.). It provides a rich agent UX, including file-edit diff cards, checkpoints, and semantic search, treating local models as primary rather than secondary options. The system is engineered for resilience, adeptly handling the unique challenges and failure modes of local models to ensure robust tool-call execution and interaction.

Quick Start & Requirements

• Install: One-line curl (macOS/Linux) or PowerShell (Windows) script.
• Prerequisites: Local model server (Ollama, LM Studio, or OpenAI-compatible), a coding model (e.g., qwen2.5-coder:7b), VS Code ≥ 1.84 or Cursor. Node.js ≥ 18 required for source builds.
• Links: Installation scripts available via GitHub.

Highlighted Details

• Offline & Private: Operates entirely offline with optional network use for specific downloads; no telemetry or cloud upsell.
• Rich Agent UX: Features file-edit diffs, checkpoints, approval cards, @-mentions, semantic codebase search, job queue, project memory, and MCP tool integration.
• Inference Fleet Management: Manages multiple local endpoints, auto-discovers models, probes capabilities, and supports automatic failover.
• Local Model Resilience: Robust handling of tool calls from various model output formats, argument repair, and whitespace-tolerant edits.
• Zero Marginal Cost: Leverages existing hardware with no per-token billing.
• Semantic Search: Local RAG powered by an int8-quantized embedding index for concept-based code retrieval.
• Tab Autocomplete: Integrates local FIM models for inline code completions.
• Browser Automation: Headless browser control for agent interaction with web pages.

Maintenance & Community

The README points to the GitHub repository for issues and updates. No specific community channels (e.g., Discord, Slack) or contributor details are listed.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The project is actively developed (v0.29.0) with features like advanced semantic indexing and marketplace publishing still on the roadmap. Performance is inherently tied to the user's local hardware and model choices. It requires a running local model server as a prerequisite.