Changelog

Scope

• Fixed WebSocket status flapping caused by Render's idle timeout dropping connections faster than the extension's reconnect could recover. Added per-model heartbeats, stale-model reaper, and debounced reconnect to eliminate online/offline cycling.
• Added user profile page with ImageKit-powered avatar upload, workspace logo support, and CDN-backed image delivery.
• Replaced the dashboard loading state with an animated PullO loader across all pages.
• Added SSE streaming for the model test endpoint with ThinkingIndicator and StreamingText components.
• Refreshed dashboard UI with glassmorphism panels, ambient background artwork, and a unified design system across models, API keys, logs, and analytics pages.
• Added user settings page and custom 404 page.
• Reduced extension heartbeat interval from 30s to 10s for faster disconnect detection.

• Integrated Corsair SDK as a Universal MCP Gateway, enabling workspace-scoped MCP server management and tool injection into the chat completion loop. Supported integrations: GitHub, Slack, Gmail, and Linear.
• Launched the documentation site (pullo-docs.vercel.app) with 15 MDX pages covering Quickstart, API Reference, Architecture, MCP, Tools, Teams, CLI, Extension, FAQ, Troubleshooting, SDKs, and Changelog.
• Implemented RBAC (Owner/Admin/Member) enforced at the API layer for all workspace operations.
• Replaced CORS-blocked fetch-based web search with Chrome offscreen document API, including bot-detection validation and tab-based fallback.
• Added Sentry error tracking for both backend (FastAPI) and frontend (Next.js) with custom metrics for API keys and page views.
• Added Recharts-based analytics dashboard with requests, latency, tokens, and hourly/daily/weekly views.
• Replaced SMTP with Resend for team invitation emails, including professional HTML email templates.
• Added Privacy Policy and Terms & Conditions pages.

• Initial release of the Chrome Extension v2.0 with Ollama, LM Studio, and Custom URL provider management. Included multi-tab popup UI and a tool execution loop supporting up to 10 iterations with browser runtime tools (web search, page reader, screenshot, calculator, and more).
• Implemented workspace-scoped architecture with database migrations for models, devices, and API keys. Added device registration flow (ADQ-002/003).
• Deployed Redis infrastructure for rate limiting (RPM/RPD), request queuing, heartbeat monitoring, and realtime counter services via Upstash Redis.
• Launched OpenAI-compatible API endpoints (`/v1/chat/completions`, `/v1/embeddings`) with pull-based WebSocket transport, API key authentication (SHA-256 hash), and per-key policy enforcement.
• Built the Next.js 14 dashboard with Supabase auth (email/password + Google OAuth), sidebar layout, and live backend metrics for models, API keys, and workspace members.
• Published the interactive 3D landing page with Three.js laptop model, network particles, and terminal animation.
• Set up GitHub Actions CI/CD pipelines for backend (lint + test), frontend (lint + build), and extension (lint + package).