EN
ES
PT
tab-agent
Choose where your model runs — the provider credential is never exposed:
In this tab (WebGPU)
My endpoint
Operator proxy
Qwen2.5-1.5B — fast (~950MB, cached)
Qwen2.5-3B — better (~1.9GB)
Hermes-3-Llama-3.2-3B (~1.9GB)
Endpoint
Model
API key (stays in this tab)
Proxy URL (Worker holds the secret)
Model
▸ Advanced: tools origin & shared memory
Tools origin (hash-verified)
Memory repo (cq-git, owner/name)
GitHub PAT for writes (optional)
Start agent
Reconfigure
Run
▸ Show verification & tool trace