SmartCursorX is an optimized fork designed to run secure local development environments with robust AI integrations, zero plain-text credential leakages, and dynamic policy steering.
All API keys and tokens are securely encrypted using OS-level credential management (DPAPI on Windows via safeStorage), preventing any plaintext leakage in database tables or logs.
Enforce cognitive steering and dynamic prompt tailors directly at model boundaries. Restrict sensitive local system tools based on explicit user taxonomy domains.
The internal administration REST endpoint is strictly bound to local loopback (127.0.0.1) and enforces strict Host header filters, neutralizing DNS rebinding attacks completely.
Fully scrutinized codebase using deterministic regex scanners to ensure no OpenAI keys, JWTs, or private keys remain inside source codes or logs.
Monaco Editor Renderer process executes inside isolated contexts, communicating only via predefined, strongly whitelisted Electron IPC events.
File edits and snapshot loops validate targets against active workspaces, aborting any operation attempting relative directory traversal (../).
Configure OpenAI, Anthropic, Gemini, OpenRouter, Ollama, LiteLLM, local models, and fine-tuned providers. API key security with DPAP storage.
Run quantized GGUF models locally with bundled llama-server. HuggingFace search, model download, server setup, and troubleshooting.
Train custom LoRA adapters using local hardware. Hardware detection, model recommendations, dataset preparation, micro-benchmarking, and inference.
Pluggable agent system with cognitive steering, taxonomy-aware context assembly, tool permissions, and plan-execute-review loop architecture.
Track token usage and estimated costs across all AI providers. Token budget monitoring, model statistics, and cost estimation with configurable pricing.
Build and package the production installer. Run from the repository root:
The installer is generated in the local releases directory:
Push commits and upload the installer as a release asset under tag v0.0.4-alpha.