KeylessAI is a drop-in OpenAI-compatible LLM endpoint that requires no API key, no signup, and no backend. Point any OpenAI client at https://keylessai.thryx.workers.dev/v1 with any non-empty string as the API key, and existing OpenAI code works unchanged.

How do I use KeylessAI with Aider?

Set environment variables: OPENAI_API_BASE=https://keylessai.thryx.workers.dev/v1, OPENAI_API_KEY=not-needed. Then run: aider --model openai/openai-fast

How do I use KeylessAI with Cline, Roo Code, or Continue.dev?

All three accept a custom OpenAI base URL. In VS Code settings, set the provider to openai, baseUrl to https://keylessai.thryx.workers.dev/v1, apiKey to not-needed, and model to openai-fast.

How do I use KeylessAI with LangChain?

Use langchain_openai.ChatOpenAI with base_url set to https://keylessai.thryx.workers.dev/v1 and api_key set to any non-empty string. Model: openai-fast.

Does KeylessAI work with Claude Code?

Not directly — Claude Code speaks Anthropic's Messages API, not OpenAI chat-completions. Bridge through LiteLLM proxy: install litellm[proxy], configure a model_list that maps Anthropic model names to openai/openai-fast at api_base https://keylessai.thryx.workers.dev/v1, run litellm --config config.yaml --port 4000, then set ANTHROPIC_BASE_URL=http://localhost:4000. See examples/claude-code-bridge.md in the repo.

What model does KeylessAI provide?

The anonymous tier provides openai-fast, which is GPT-OSS 20B served by Pollinations.ai (and aliased to gpt-4o, gpt-4o-mini, claude-3-5-sonnet-latest, and 10 other popular model names). It supports tool calling and emits reasoning tokens. Good for prototyping, boilerplate, classification, and small agent loops — not a replacement for GPT-4 or Claude 3.5 on hard reasoning.

Is KeylessAI really free?

Yes. There are no API keys, no signup, no credit card, no email required. The project is MIT-licensed. The public Worker is hosted on Cloudflare's free tier at https://keylessai.thryx.workers.dev/v1. Inference is served by four keyless no-auth upstreams: Pollinations SSE, Pollinations GET, ApiAirforce, and Yqcloud. No user compute required.

What happens if one of the backends is rate-limited or down?

KeylessAI's router automatically fails over to the next provider in the pool, re-ranked every request by live success rate and TTFB. A single-flight queue serializes parallel callers under Pollinations' 1-concurrent-per-IP limit, a circuit breaker skips providers after five consecutive failures (30 s cooldown), a per-IP rate limiter protects the Worker, and promo-URL / deprecation-notice injections are detected and retried with backoff up to three times before switching providers.

KeylessAI — free LLM inference,
with zero API keys.

Chat below, or drop https://keylessai.thryx.workers.dev/v1 into your OPENAI_API_BASE and use any string as the key. Your existing OpenAI code, Aider, Cline, Codex, LangChain — all work. Zero signup. Zero compute on your end. Click </> API for exact snippets.

KeylessAI — free LLM inference,
with zero API keys.

What is this?

How is it free?

Want to use these in your code?

KeylessAI — free LLM inference, with zero API keys.

KeylessAI — free LLM inference,
with zero API keys.