🤖 Free AI instead of ChatGPT and Claude — 34 providers, 635 endpoints, one key

[SRZ]

One key for every free AI model

Free AI is not scarce any more — keys are. Every provider hands out a little, and pooled behind one endpoint those little bits become something you can build on all day.

What that one key gives you

• chat with the frontier models
• coding inside opencode, Cline, Aider or Continue
• agents that run your scripts and tools

The pools — pick one and you are set

• FreeLLMAPI ★38k · MIT — 34 providers, 635 free endpoints, 7.4B tokens a month behind a single /v1
• freellmpool ★128 — 23 providers, 178 routes, starts without a key
• LiteLLM ★62k — self-hosted proxy covering 140+ providers
• free-router ★478 · CliRelay ★1k — model ranking, and CLI logins bridged into one endpoint

When a key runs out — and it will

• new-api ★50k — hold every key you own, rotate and retry automatically
• LiteLLM router — fallbacks, retries and cooldowns in one config
• Cloudflare AI Gateway · OpenRouter routing — hosted fallbacks, no server to keep alive

Keys with no card and no phone — Google AI Studio · Groq · Cerebras · SambaNova · Cloudflare Workers AI · NVIDIA NIM · Hugging Face · GLM on z.ai · OpenRouter · and DuckDuckGo AI asks for nothing at all

Agents, all free — opencode ★219k · Gemini CLI ★107k · Cline ★71k · Aider ★50k · Continue ★36k — or keep it offline with Ollama ★184k and Open WebUI ★156k

Three moves to start — install opencode → paste the FreeLLMAPI key → put new-api in front so a dead key never reaches you

Honest note — free tiers are rate-limited, and the rotation layer is how people work around it. Worth reading each provider’s data policy before pasting company code

Directories that stay updated — itsfree.ai · freeinference.dev · freellmapi.co

[ai_daily_tools]

Great roundup, thanks for putting this together. The rotation layer is the part most people skip and then complain when a key dies mid-session — learned that the hard way. I’m a bit wary of pooled keys for anything with personal data though, since you never know who’s logging what on the other end. For quick experiments and throwaway prompts this looks like a solid starting point.

[Grace2077]

Solid roundup — the rotation layer is the part most people skip and then regret. One practical addition: configure fallbacks per model, not per provider, because free tiers usually die one model at a time, and a provider-level fallback will happily route you to another dead model. Also keep one local model as the last fallback in the chain, so a total outage degrades you to slow instead of dead. And take the pooled-quota figures with a grain of salt — free pools get throttled fastest exactly when everyone reads a post like this, so spread heavy jobs across at least two pools from day one.