v3.0.0 — Gemini 4 Argon Frontier Intelligence & Deep Thinking Engine

🚀 Gemini 4 Argon Frontier Intelligence & Deep Thinking Release (v3.0.0)

We are thrilled to announce v3.0.0, unlocking programmatic access to Google's next-generation Gemini 4 Argon frontier intelligence model with customizable Deep Thinking controls, alongside Gemini 3.8 Flash as the ultra-fast default engine.

🌟 What's New in v3.0.0:

• 🧠 Google Gemini 4 Argon Native Support:
  • Programmatic access to Google's frontier Argon model (gemini-4-argon).
  • Reverse-engineered JSPB protocol header injection unlocking deep thinking depth:
    • argon-high (Level 2): Thorough step-by-step reasoning for difficult riddles, competitive coding, and complex logic deduction.
    • argon-medium (Level 4): Balanced reasoning for technical explanations and debugging.
    • argon-low (Level 1): Rapid reasoning with verification.
• ⚡ Gemini 3.8 Flash as Default (is_default: true):
  • Maintained as the ultra-fast default model for zero-latency everyday completions.
• 🛠️ Pure Go Output Arranger:
  • 100% pure Go Output Arranger engine for zero-dependency native OpenAI tool & function calling.
• 🌐 Multi-Account Routing & Resilient Worker Pool:
  • Multi-account routing (/u/N/) support ensuring 100% cookie auth alignment with Google accounts.
  • Automatic session health watchdog with zero-duplicate worker management.
• 📊 Universal Thinking Controls:
  • Added thinking controls across the entire Gemini family (flash-high, flash-medium, flash-low, pro-high).

v2.2.0 — Autonomous Network Discovery & Gemini 3.8 Flash Standard

Release Highlights

🌐 Autonomous Zero-Config Local Network Discovery

• Sub-300ms Multi-Tier Discovery Engine: The companion Chrome extension automatically scans localhost, standard mDNS hostnames, and local subnets in concurrent batches to locate the central Go API server without requiring manual IP configuration.
• Multi-Device Cookie Aggregation: Deploy the companion extension across multiple workstations on your local network to aggregate Google accounts into a centralized high-capacity worker pool.
• Self-Healing Connection Recovery: Caches working server endpoints for instantaneous reconnects and automatically triggers subnet discovery if the server host IP changes.

⚡ Strict Gemini 3.8 Flash Standardization

• Enforced Model Uniformity: Standardized all chat endpoints, streaming handlers, and multi-agent queue workers strictly on gemini-3.8-flash.
• Accurate Token Accounting: Real-time token usage estimation using OpenAI cl100k_base BPE tokenizer.

🛠️ Interactive Developer CLI & Diagnostics

• CLI Developer Manual: Access comprehensive operational guides directly via GET /help, GET /, or ./goapi --help.
• Live Worker Metrics: Real-time inspection of active workers, in-flight concurrency, and individual account health via GET /v1/workers.

📦 Package & Component Updates

• Upgraded Chrome Extension manifest to v2.2.0.
• Upgraded Model Context Protocol (MCP) server definition to v2.2.0.
• Refactored project documentation to enterprise open-source standards.

v2.1.0 — Multi-Agent Parallel Worker Queue Engine

🚀 v2.1.0 — Multi-Agent Parallel Worker Queue Engine (queue.go)

👥 Parallel Multi-Account Scaling (1 to 100+ Accounts)

• Dynamic Worker Pool: Discovers all Google accounts in cookies/ and initializes concurrent workers.
• True Multi-Agent Parallelism: 10, 50, or 100 agents execute queries simultaneously across distinct accounts.
• Least-Busy Load Balancing: Routes requests to the worker with the lowest in-flight count.
• Sticky Session Affinity: Preserves multi-turn conversation context (c_...) using user, X-Agent-ID, or X-Conversation-ID.
• Circuit Breaker (429 Protection): Auto-quarantines rate-limited accounts for 45s while seamlessly serving requests from healthy workers.
• Live Cluster Monitoring: Real-time stats at GET /v1/workers and GET /api/pool.
• Safe Streaming SSE: ExecuteQueuedStream with guaranteed worker resource release.

v2.0.0 — Docker + Reactive Cookie Architecture

🚀 v2.0.0 — Docker/OrbStack + Reactive Cookie Sync

🐳 Docker / OrbStack

• Fully self-contained Docker image (no host folder mounts)
• Multi-stage Alpine build (~25MB final image)
• restart: unless-stopped — auto-starts on Mac reboot
• Ports 8001 (API) and 9226 (WebSocket cookie bridge)

🔄 Reactive Cookie Architecture

• Only-On-Expiry sync — zero proactive polling
• ExecuteWithFailover() triggers cookie refresh on 401 Unauthorized
• Passive 30-minute health logger
• Dual resilience: WebSocket primary + POST /api/sync-cookies HTTP fallback

🧩 Chrome Extension

• 📋 Copy Cookies button with clipboard animation
• Host permissions for 127.0.0.1 & localhost (Docker-compatible)
• Real-time passive cookie listener on Google session token rotation

⚡ Performance & Stability

• QUIC timeout: 60s → 10s for fast fallback
• Fixed 401 bug from duplicate Cookie header in HTTP/2 (tls-client CookieJar conflict)
• CORS middleware added for extension fetch() compatibility
• BPE tokenizer (cl100k_base) — 99.5% token accuracy
• Dual analytics export: Excel (.xlsx) + JSON auto-generated on every request

📦 Cookie Lifetime (Real Test Data)

Metric	Result
Unbroken session	1h 24m 30s (76 requests, 100% pass)
Total tested lifespan	2h 05m 47s (99 queries)
Expiry signal	HTTP 401 on StreamGenerate
Recovery	Automatic via extension WebSocket push

---

Full documentation: README.md

Free Gemini API & Engine v1.0.0

🚀 Initial release with Gemini 3.7 Flash, Needle 2 SLM, and SQLite Persistence.