llm8: The Free-Tier AI Pool

llm8 is a local rotating proxy for free-tier AI accounts (Gemini, Anthropic, Kiro, etc.) that pools credits into a single OpenAI-compatible endpoint.

Stop manually swapping API keys when you hit a rate limit. Point your IDE at llm8, add your accounts, and enjoy a much larger, combined quota that rotates automatically.

---

🚀 Key Features

• One URL for everything: Point VSCode + Continue/Cline/Roo at http://127.0.0.1:7842/v1.
• Automatic Rotation: When one account hits a rate limit (429), llm8 instantly switches to the next available one.
• Cross-Vendor Intelligence: Ask for claude-3-5-sonnet and llm8 will find it in your Anthropic, Kiro, or Antigravity accounts automatically.
• Smart Cooldowns: Accounts are automatically "parked" until their quota resets.
• Local & Private: No tokens or data leave your machine. Everything is handled by a local daemon.

---

🛠 Supported Vendors

Vendor	Account Type	Credits (approx)
Google Gemini	AI Studio / Code Assist / Vertex	Generous Free Tier
Anthropic	API Keys	Standard Quota
Antigravity	Google Browser-OAuth	High-Throughput Gemini/Claude
Kiro (Amazon Q)	AWS Builder ID	50 credits/mo per account

---

🏁 Get Started in 3 Minutes

1. Build and Install

git clone https://github.com/ai-agent-lead/llm8.git
cd llm8
make install

2. Start the Proxy

llm8 start &

3. Add an Account

llm8 accounts add aistudio --api-key "YOUR_KEY"
# OR
llm8 accounts add anthropic --key "sk-ant-..."

4. Point Your IDE at the Pool

In your IDE extension (Continue, Cline, etc.):

• Base URL: http://127.0.0.1:7842/v1
• API Key: pool
• Model: claude-3-5-sonnet-20241022 or gemini-1.5-pro

---

📖 Documentation

• User Guide: Full setup instructions for all vendors and IDEs. Start here.
• Account Safety: How to avoid bans and use "burner" accounts. Read this.
• CLI Reference: Commands for stats, usage, and pool management.

---

🔧 Technical Details (for Developers)

llm8 is written in Go and uses a local SQLite database to track usage and quota state. It translates OpenAI-shaped requests into vendor-specific protocols (SSE, Custom JSON, etc.) and back again.

For architecture details, see:

• Architecture Overview
• Domain Glossary (CONTEXT.md)
• Design Decisions (ADRs)

---

⚠️ Important: Account Safety & ToS Posture

When using Kiro (AWS Builder ID) accounts, you must follow strict safety guidelines to avoid suspension:

• Use Burner Accounts: Always use a dedicated burner Builder ID.
• No Cross-Tool Usage: Do NOT run kiro-gateway or the Kiro IDE extension against the same account used in llm8.
• Personal Use Only: This is a self-host tool for personal-use; do not re-host publicly.
• Known Issues: We cite documented 2026 suspension waves (#5077, #4963, #6097, #6632, #6669) as evidence that AWS monitors for unusual activity.
• No System-Wide Unban: Unlike some vendors, Kiro has no system-wide unban precedent; suspensions are permanent.
• Fingerprint Safety: llm8 uses a unique per-account fingerprint to minimize detection.

---

License

TBD — See LICENSE (coming soon).