A low-latency, context-aware AI tab-completion engine for Visual Studio Code. GhostContext features a high-performance inference architecture supporting BYOK (FIM-compatible endpoints), telemetry-free local LLM execution, and granular AST-driven context gathering.
- Bring Your Own Key (BYOK): Directly integrate with any standard API endpoint supporting Fill-In-The-Middle (FIM) models (e.g., DeepSeek-Coder-V2, Codestral).
- Local Execution Engine: Native support for local inference via Ollama or LM Studio, ensuring zero telemetry and absolute codebase privacy.
- Managed Tier: Built-in integration for a managed inference proxy, allowing seamless onboarding and testing via free compute credits.
- Sub-200ms Latency: Optimized request debouncing and context pruning to ensure rapid ghost text rendering without rate-limit exhaustion.
The project is structured to separate editor UI from context processing:
- Client Extension: Built in TypeScript utilizing the VS Code
InlineCompletionItemProviderAPI for non-blocking UI rendering. - Context Engine: Parses the active document and neighboring tabs to construct optimal prompt structures (Prefix + Suffix) for FIM models.
- Inference Proxy (Optional Backend): Compatible with high-performance network runtimes (ike Bun or Go) to handle rate-limiting, authentication, and token management for the managed tier.
- Clone the repository:
git clone https://github.com/SAYOUNCDR/GhostContext.git