Free web chat UI + API proxy fronting OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba, DeepInfra, and Meta through one endpoint
Access model
No accounts required — a browser-bound client ID handled rate limiting; optional email sign-in unlocked persistent xav- keys with no rate limit
Compatibility layer
OpenAI-compatible (/v1/chat/completions), Anthropic-compatible (/v1/messages), and a legacy custom schema — drop-in for SDKs, CLIs, and coding agents
Rate limit
120 req/min per browser client ID (not per IP), sliding window; verified users bypassed it entirely
Operator
Anonymous individual, self-described "hobby project" — no company, no legal entity, no jurisdiction of record
Funding
Entirely out of pocket across all eight providers — no ads, no donations, no accounts to bill
Effective docs
Privacy statement effective June 5, 2026, last revised June 20, 2026
// Why it's retired
This ran on real, official provider API keys that the operator paid for personally,
proxied through a private gateway (a rented VPS, not a serverless function) sitting between
Cloudflare Workers and each provider. The operator's own privacy statement disclosed the exact
risks this created, rather than hiding them.
"There is no registered agent for service of process, no data protection officer, and no formal complaint handling procedure."
"If monthly costs exceed what the operator can afford, the service will be taken down, possibly without notice."
"Any interaction with this service ... could theoretically deliver malware if the upstream gateway ... were compromised or malicious."
Legal gray area by the operator's own admission — proxying paid API access to anonymous users likely violates most providers' terms of service; access could be revoked at any time with no recourse
Unverifiable gateway — unlike the stateless Cloudflare Worker layer, the upstream gateway ran on a persistent VPS that was technically capable of logging, altering, or replaying prompt content, and the operator's no-logging claim could not be independently audited
Agent-tool exposure — the OpenAI/Anthropic-compatible endpoints were built to be dropped straight into coding agents and editors, routing prompts, source code, and tool calls through that unverifiable gateway on every request
Funding-dependent uptime — no reserve fund, no donations mechanism (deliberately), no succession plan; existence depended entirely on one person's ongoing willingness to pay
Zero track record — the service had no independent audit, no uptime history, and, in the operator's own words, "nothing beyond documented claims to evaluate"
// Active endpointThis relay is dark, but the project didn't stop. A newer instance is live at a different address.