An honest chat API. Direct answers, no scaffolding, no fluff. Backed by Qwen/Qwen3.5-397B-A17B. The agent can shell, deploy CF, and hit pay.sh APIs while spending its agentcash wallet — invisible to you, just shows up as the final answer.
curl -X POST https://chat.unbrowse.ai/v1/chat/completions \
-H 'Content-Type: application/json' \
--data '{
"messages":[{"role":"user","content":"Why does the structural-prior matter? One sentence."}]
}'
1 credit ≈ 1,000 output tokens chat, or 10 credits per remote exec call, or whatever pay.sh quotes for that hop. Markup goes to infra + uptime + the agent's own wallet replenishment.
messages, stream, tools — full OpenAI surface debits credits per 1k output tokens x-trace-id header on every response
{ "command": "echo hi", "timeout_s": 60 }
debits 10 credits/call
proxies to a backend VM with shell access
returns { balance, currency: "credits" }
free-tier IPs see per-day allowance
first event: trace (id, model, mode) then OpenAI delta chunks reasoning_content carries the thinking
chat (1k out tokens) ~1 credit remote shell exec (60s) 10 credits cf pages deploy (via shell) ~30 credits · marked-up from real CF cost pay.sh paid API call quoted ± 20-100x markup agentcash x402 hop quoted ± 20-100x markup
by Foundry · agent rails via pay.sh + agentcash · payments by Stripe · hosted on Cloudflare · backed by Nebius