Skip to content

CHANGELOG.md

Latest commit

 

History

History
163 lines (140 loc) · 10.7 KB

CHANGELOG.md

File metadata and controls

163 lines (140 loc) · 10.7 KB

Changelog

All notable changes to HugBrowse will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.0.0] - 2025-07-19

Added

  • Model Management Engine (Phase 1/8 of v1.0)

    • New model_manager.rs Rust module for full model lifecycle management
    • Multi-instance model loading — run multiple models simultaneously on unique ports
    • JIT (Just-In-Time) loading support — auto-load models on first API request
    • TTL (Time-To-Live) auto-unload — idle models unload after configurable timeout (default 60min)
    • Auto-eviction — LRU JIT-loaded models evicted when memory is insufficient
    • Per-instance health checking with configurable retry limits (max 3 retries)
    • Memory usage tracking (RAM/VRAM) per loaded model instance
    • New Tauri commands: mm_load_model, mm_unload_model, mm_unload_all, mm_list_loaded_models, mm_get_model_status, mm_get_memory_usage, mm_update_config
    • New modelManager Zustand store for frontend model management state
    • New LoadedModelsPanel component showing all loaded models with status, memory, and unload controls
    • Real-time model status events via Tauri event system (model-status-changed, model-ttl-expired)

  • API Server (Phase 2/8 of v1.0)

    • New api_server.rs Rust module — embedded axum HTTP server inside Tauri app
    • OpenAI-compatible endpoints: GET /v1/models, POST /v1/chat/completions (SSE streaming + non-streaming), POST /v1/completions, POST /v1/embeddings
    • Native REST API: GET /api/v1/status, GET /api/v1/models, POST /api/v1/models/load, POST /api/v1/models/unload, POST /api/v1/chat
    • Configurable port (default 8080), start/stop via Tauri commands
    • SSE streaming proxy — streams llama-server responses directly to API clients
    • Standard OpenAI error format on all endpoints (message, type, code, requestId)
    • Request logging with in-memory log and real-time Tauri event emission
    • CORS support via tower-http
    • New Tauri commands: api_server_start, api_server_stop, api_server_status, api_server_update_config, api_server_get_log
    • New apiServer Zustand store for frontend server management state
    • New Developer page with server controls, status indicator, base URL display, and request log viewer
    • Developer navigation tab in header

  • Authentication & Security (Phase 3/8 of v1.0)

    • New auth_manager.rs Rust module — token-based API authentication system
    • SHA-256 hashed token storage — plaintext shown once on creation, never persisted
    • Encrypted persistence via tauri-plugin-store (auth-tokens.json)
    • Per-token permission scoping: inference, model_management, server_admin, downloads
    • Auth enforcement on all API server routes — returns 401/403 for invalid/insufficient tokens
    • Auth toggle — enable/disable authentication globally without deleting tokens
    • New Tauri commands: auth_create_token, auth_revoke_token, auth_delete_token, auth_list_tokens, auth_set_enabled, auth_is_enabled
    • New auth Zustand store for frontend token management state
    • New TokenManager component in Developer page — create, revoke, delete tokens with permission grid
    • Token creation displays plaintext token once with copy-to-clipboard support

  • Configuration & Presets (Phase 4/8 of v1.0)

    • New config_manager.rs Rust module — unified configuration system
    • AppSettings struct with categories: General, Models, Server, Appearance, Downloads, Advanced
    • Persistent settings to ~/.hugbrowse/config/settings.json with safe defaults
    • model.yaml parser and generator — portable model descriptors with load/inference defaults
    • Inference presets system — CRUD with JSON persistence in ~/.hugbrowse/config/presets/
    • Per-model settings with merge priority: per-model > model.yaml > preset > global defaults
    • Settings import/export support
    • Preset import/export support
    • New Tauri commands: config_get_settings, config_update_settings, config_list_presets, config_create_preset, config_update_preset, config_delete_preset, config_get_per_model, config_set_per_model, config_parse_model_yaml, config_save_model_yaml, config_export_settings, config_import_settings, config_export_preset, config_import_preset
    • New configStore and presetStore Zustand stores for frontend state management
    • New PresetManager component in Developer page — create, delete, expand presets with parameter display
    • Added serde_yaml and chrono Rust dependencies

  • Enhanced Chat System (Phase 5/8 of v1.0)

    • Chat folders — organize conversations in named folders with collapse/expand
    • Conversation search — filter sessions by title and message content
    • Preset selector in chat toolbar — apply inference presets per conversation
    • Message editing — edit messages and truncate subsequent conversation history
    • Session duplication — fork conversations with "Duplicate" action
    • Export conversations as JSON or Markdown
    • Import conversations from JSON
    • Per-conversation preset binding (presetId on ChatSession)
    • Folder CRUD (create, rename, delete) with sessions auto-unfoldered on delete
    • Enhanced SessionSidebar with search input, folder hierarchy, and duplicate button

  • MCP Integration (Phase 6/8 of v1.0)

    • New mcp_host.rs Rust module — MCP server management with JSON-RPC 2.0 over HTTP
    • Parse and persist mcp.json configuration file
    • MCP server lifecycle: add, remove, connect, discover tools
    • Tool execution with configurable timeout (default 30s)
    • Tool approval mode toggle
    • Allowed tools filter per server
    • New Tauri commands: mcp_list_servers, mcp_add_server, mcp_remove_server, mcp_connect_server, mcp_call_tool, mcp_get_config, mcp_set_approval_mode, mcp_get_approval_mode
    • New McpManager component in Developer page — add/remove servers, connect, view tools
    • Server status indicators (connected/disconnected/error) with tool count display

  • CLI Interface (Phase 7/8 of v1.0)

    • New src-cli/ standalone Rust crate — hb command-line interface
    • hb chat — interactive chat with loaded models via API server
    • hb ls — list downloaded local models
    • hb ps — list currently loaded/running models
    • hb load — load a model by path with optional context length
    • hb unload — unload a running model by ID
    • hb server status — check API server status
    • HTTP client module communicating with HugBrowse API server (OpenAI-compatible + native endpoints)
    • Configurable server URL via --url flag (default http://127.0.0.1:8080)
    • Built with clap 4 for argument parsing, reqwest for HTTP, tokio async runtime

  • Advanced Features (Phase 8/8 — v1.0.0 Release)

    • Speculative decoding support — draftModel option in LoadOptions, --model-draft flag passed to llama-server
    • Per-model inference metrics — total tokens generated, average tokens/sec, running average calculation
    • New mm_record_metrics Tauri command for recording inference performance
    • HugLink cross-device module — huglink.rs with device discovery data model, enable/disable, preferred device, rename
    • New Tauri commands: huglink_set_enabled, huglink_is_enabled, huglink_list_devices, huglink_set_preferred, huglink_rename, huglink_status
    • Plugin system — plugin_host.rs with plugin manifest parsing (plugin.json), lifecycle management (install/enable/disable/uninstall), directory scanning, sandboxed plugin configuration
    • New Tauri commands: plugin_list, plugin_enable, plugin_disable, plugin_uninstall, plugin_rescan, plugin_get, plugin_set_config
    • Enhanced RAG engine — configurable chunking strategies (fixed-size, sentence-boundary, semantic/paragraph), configurable chunk/overlap sizes, embedding model selection placeholder, setRagConfig/getRagConfig API
    • HugLink Zustand store (huglink.ts) and HugLinkPanel component in Developer page
    • Plugin Zustand store (pluginStore.ts) and PluginManager component in Developer page
    • Inference MetricsDashboard component — aggregate stats (loaded models, total requests, total tokens, avg t/s), memory usage bars, per-model metrics with speculative decoding indicator
    • Enhanced status bar — loaded model count, API server status indicator with port, version badge updated to v1.0.0
    • Version bumped to 1.0.0 in package.json and Cargo.toml

[0.3.0] - 2026-03-09

Changed

  • Version bump to v0.3.0 for automated release pipeline validation
  • Auto-updater flow: installed clients check latest.json on launch → show update notification → one-click install

[0.2.0] - 2026-03-09

Added

  • Cloud Offload — run inference on remote servers instead of (or alongside) local hardware
    • Compute Backend abstraction: Local Sidecar, HuggingFace Endpoints, Custom URL
    • Backend selector dropdown in chat interface with status indicators and latency
    • Add/remove/switch backends from Settings → Compute Backends
    • Connection testing with latency measurement for custom endpoints
    • One-click "Deploy to Cloud" from any model detail page via HuggingFace Inference Endpoints
    • HF Endpoint lifecycle management: deploy, check status, pause, resume, delete
    • Streaming chat proxy through Tauri backend (solves CORS, centralizes auth)
    • Backend tracking on chat messages — see which backend generated each response
    • Persistent backend configuration across app restarts
    • Secure API key storage for remote endpoints

[0.1.0] - 2026-03-08

Added

  • Initial release of HugBrowse
  • Model browser with search and filters (pipeline type, library, sort order)
  • Hardware tier detection (Entry / Mid / High / Ultra) and compatibility scoring
  • Download manager with pause, resume, cancel and SHA-256 integrity verification
  • Local inference engine via llama-server sidecar with GPU auto-detection (CUDA / Metal / Vulkan)
  • Chat interface with markdown rendering and streaming responses
  • Resource monitor for CPU, RAM, GPU, VRAM, and disk usage
  • Community marketplace for browsing, publishing, and reviewing extensions
  • Plugin system architecture for community-contributed extensions
  • Auto-updater integrated with GitHub Releases
  • Deep-link support via hugbrowse:// protocol for one-click model imports
  • System tray with quick actions
  • Onboarding wizard for first-time setup
  • RAG document attachment support (PDF, DOCX, plain text)
  • Privacy-first design — all inference and data stays local
  • GGUF file association for direct model opening
  • MSI and NSIS installer targets for Windows

Fixed

  • MSVC toolchain detection for Windows builds
  • Shell plugin v2 configuration compatibility