Teale ∙ manifesto | for OpenClaw | for Hermes | for Developers
Your agents don't need a centralized API.
Hermes speaks OpenAI. Teale is OpenAI-compatible, backed by the world's idle laptops. Tool use, chains, memory, streaming — all keep working.
Reliable supply
Most devices on the network run a unified Hermes-3-Llama 3.1 8B. Your agent's multi-step calls route across whichever laptops are free — consistent model, consistent behavior between steps.
Performance
| Model | TTFT | TPS | Credits / 1M |
|---|---|---|---|
| Hermes-3-Llama 3.1 8B | ~200ms | ~45 | 800 |
| Qwen 3 32B | ~350ms | ~18 | 3,200 |
| Llama 4 Scout 109B MoE | ~500ms | ~12 | 10,900 |
Illustrative figures. Live metrics stream from gateway.teale.com.
Switch your base URL
Point Hermes at Teale. Keep every other setting.
export OPENAI_BASE_URL="https://gateway.teale.com/v1"
export OPENAI_API_KEY="tok_..."
Early tester builds — you'll need to bypass your OS security prompt on first launch (Gatekeeper on Mac, SmartScreen on Windows). Linux ships today as an unsigned x86_64 tarball. The apps aren't signed yet.