Gemini Local UI (gemini_local_ui)

High-performance local web client, reverse-engineered gemini.google.com session bridge, SQLite FTS5 conversation store with DAG branching, and OpenAI-compatible local API server (/v1/chat/completions, /v1/models).

Key Capabilities

1. Authorized Gemini Web-Session Bridge (backend/app/bridge/)

   • Authenticates against gemini.google.com using your authorized session cookies (__Secure-1PSID, __Secure-1PSIDTS) or automatic local Chromium/Chrome/Edge/Brave profile extraction.
   • Extracts SNlM0e CSRF tokens, executes batchexecute RPCs (otAQ7b, MaZiqc, hNvQHb, CNgdBe), parses length-prefixed StreamGenerate chunks (with incremental UTF-8 frame decoding and thinking-trace separation), uploads multimodal files via content-push.googleapis.com, and rotates session cookies via RotateCookies.
   • Encrypts stored session credentials locally using native Windows DPAPI (CryptProtectData / CryptUnprotectData in backend/app/storage/vault.py) with AES-256-GCM fallback.

2. Lag-Free Virtualized Local Web UI (frontend/src/)

   • Custom viewport virtualizer (VirtualChatList.tsx) with O(log N) binary-search offset indexing, synchronous layout measurement + singleton ResizeObserver dynamic height propagation, and scroll anchoring.
   • Off-main-thread Web Worker syntax highlighting (highlight.worker.ts) and Shadow DOM style/token encapsulation (CodeBlock.tsx, MathBlock.tsx), maintaining ~83–106 total document DOM nodes and 0 main-thread long tasks (>50ms) even on 500–1,050 message threads with 100+ code blocks and LaTeX formulas.

3. Local SQLite WAL Storage, DAG Branching & Instant FTS5 Search (backend/app/storage/)

   • Multi-connection reader pool + serialized write lock over SQLite WAL mode.
   • Full-text search (FTS5 with BM25 ranking, snippet highlighting, single-flight request coalescing, and generation-epoch cache invalidation) delivering ~19ms P99 latency across 1,000+ conversations under 100-client concurrency.
   • Tree-structured message branching (parent_id DAG), sibling navigation (< 1/2 >), message editing with fork regeneration, and Markdown/JSON conversation export.

4. OpenAI-Compatible Local API Proxy (backend/app/openai/)

   • Exposes GET /v1/models and POST /v1/chat/completions (both stream: true SSE and stream: false, plus base64/URL vision attachments) routed through your authorized Gemini account.

5. Google Flow Educational Video Studio (backend/app/bridge/flow.py, backend/app/api/flow.py, FlowStudioDrawer.tsx)

   • Integrates with https://labs.google/fx/tools/flow (Veo 3.1) to synthesize multi-scene educational storyboards, per-scene visual prompts, voiceover narration scripts, camera motion directives, synchronized WebVTT (.vtt) / SubRip (.srt) subtitle tracks, and assembled .mp4 video lessons.

6. AGY Mission Control — 4-Tier Golden Stack Hub (backend/app/bridge/mission_control.py, backend/app/api/mission_control.py, MissionControlDrawer.tsx)

   • Unites the 4-Tier (L1 -> L2 -> L3 -> L4) ecosystem inside a single real-time command center:
     • P10–P50 Golden Stack Bus: Single-process priority hook multiplexer (0.06ms average 5-slot in-memory dispatch) with re-entrancy guard, P30 Agile V Self-Grading Bias enforcement, and P50 5-Gate OSS pre-push auditing.
     • LSRPC Plugin Manager: Context-safe plugin discovery and live config.json state toggling keyed by install directory (dir).
     • ISO 9001:2015 Gate Controller: Finite-state progression (SPEC -> TEST_DESIGN -> SYNTHESIS -> VERIFICATION -> RELEASE) with locked test SHA-256 tamper detection.
     • Brain Memory & Child Testbench Telemetry: Cross-session artifact search ([label](conversation://<id>)) and live ingestion of child_session_reports.jsonl handoffs.

---

Quick Start

One-Command Launch (Windows PowerShell)

.\start.ps1

• Backend API & OpenAI Proxy: http://127.0.0.1:8000
• Local Web UI: http://127.0.0.1:5173

Manual Launch

1. Start Backend Server:

   python -m uvicorn backend.app.main:app --host 127.0.0.1 --port 8000

2. Start Frontend UI:

   cd frontend
   npm run dev -- --host 127.0.0.1 --port 5173

Configuring Your Gemini Account Session

1. Open http://127.0.0.1:5173 and click ● Auth (top right) to open the Session & Credentials modal.
2. Either click Auto-Extract from Browser (extracts cookies from local Chrome/Edge/Brave profile) or paste your __Secure-1PSID and __Secure-1PSIDTS cookies from gemini.google.com DevTools (Application -> Cookies).
3. To use with external IDE plugins or OpenAI SDK clients:
   • Base URL: http://127.0.0.1:8000/v1
   • API Key: local-gemini (any non-empty string)
   • Models: gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash-thinking

---

Verification & Benchmarks

• 4-Tier Master E2E Suite (383 tests):

  python tests/e2e/runner.py

• Backend Integration Suite:

  python -m pytest backend/tests -v

• Playwright Headless UI & 100Hz Streaming Benchmarks:

  cd frontend
  npx playwright test

Related Projects

• iso9000-agentops — ISO 9001:2015 / Agile V Multi-Agent Quality Pipeline & Benchmark Suite
• agy-lsrpc-bridge — Language Server ConnectRPC Bridge, Plugin Validator & Scaffolder
• agy-golden-stack — Unified In-Process P10..P50 AgentOps & Hook Multiplexer Bus