0.0.2

Added

• Per-workspace conversations (existing global conversations migrate automatically)
• GGUF models auto-load on first message with a "loading model" card in chat
• llama.cpp server uses random free ports with retry on bind failure

Changed

• Composer dropdowns (model picker, mode menu) now position themselves within the viewport and work in edit mode
• All composers share one selected model and mode
• Auto model selection hidden for now; first enabled model is the default

Fixed

• Production error: Cannot find package '@huggingface/hub' (runtime deps now resolved via file URLs)

Removed

• MCP tool marketplace

Initial release

• Agent chat sidebar with multi-turn conversations and streaming responses
• Tool suite: file read/write/edit, glob/grep search, shell commands, web search/fetch
• Local model providers: Ollama and llama.cpp, plus OAuth-based cloud providers
• Semantic codebase index for meaning-based search
• MCP (Model Context Protocol) client with external server support
• Approval policy engine with allow/ask/deny rules per tool (shell, edits, web, MCP)
• Inline diff review for AI-proposed edits
• Context mentions, workspace context, and custom rules/hooks
• Settings panel (React webview) for models, features, and approval configuration
• Ctrl+L / Cmd+L to add editor selection to chat