@chirag127/oriz-ai-providers

Thin wrapper around 20 free LLM APIs with priority-based fallback. OpenAI SDK-compatible.

Install

pnpm add @chirag127/oriz-ai-providers

Quick example

import { ai } from '@chirag127/oriz-ai-providers';

// Single-prompt completion
const r = await ai.complete({ prompt: 'Rewrite for Twitter: ...', maxTokens: 280 });
console.log(r.text, '— from', r.provider, r.model);

// Multi-turn chat
const r2 = await ai.chat([
  { role: 'system', content: 'You are concise.' },
  { role: 'user', content: 'What is Cerebras?' },
]);

// Embeddings
const e = await ai.embed('hello world');

Providers

| # | Provider | Anon? | Notes | |---|---|---|---| | 1 | OVHcloud AI Endpoints | yes | 2 RPM/IP, EU-hosted | | 2 | LLM7.io | yes | 30 RPM/IP, 30+ models | | 3 | Pollinations | yes | gpt-oss-20b | | 4 | Cerebras | no | 30 RPM + 1M TPD, ultra-fast | | 5 | Groq | no | 30 RPM + 1K RPD, llama-3.3-70b | | 6 | NVIDIA NIM | no | 40 RPM, phone verify | | 7 | Google AI Studio | no | Gemini 2.5 | | 8 | Cohere | no | Command A+ | | 9 | GitHub Models | no | GPT-4.1 / o4-mini | | 10 | Cloudflare Workers AI | no | 10K neurons/day | | 11 | HuggingFace Router | no | Thousands of models | | 12 | Mistral La Plateforme | no | Mistral Medium 3.5 / Codestral | | 13 | SambaNova | no | DeepSeek V3.1 + Llama 3.3 70B | | 14 | OpenRouter | no | 20 RPM + 200 RPD :free | | 15 | Z.AI (Zhipu) | no | GLM-4.7-Flash | | 16 | SiliconFlow | no | Qwen3-8B | | 17 | Aion Labs | no | 15 RPM + 20K TPD | | 18 | Ollama Cloud | no | NOT OpenAI-compatible (skipped in v0.1) | | 19 | ModelScope | no | 2K RPD, Alibaba real-name | | 20 | Kilo Code | no | Auto-router free models |

Priority chain

Default order (highest first) — comes from priority.json in the data repo:

ovhcloud → llm7 → pollinations → cerebras → groq → nvidia-nim →
openrouter → google-aistudio → cloudflare-workers-ai → github-models →
sambanova → huggingface → mistral → cohere → zai → siliconflow →
modelscope → aion-labs → kilo-code → ollama-cloud

Override per-call: ai.complete({ prompt, preferProvider: 'cerebras' }).

Task-tuned chains: task: 'highVolume' | 'reasoning' | 'vision' | 'code'.

Environment variables

Each provider has a single env var (see env-vars.json):

OVHCLOUD_AI_KEY            (optional — anonymous works)
LLM7_API_KEY               (optional — anonymous works)
CEREBRAS_API_KEY
GROQ_API_KEY
NVIDIA_NIM_KEY
GOOGLE_AI_STUDIO_KEY
COHERE_API_KEY
GITHUB_MODELS_TOKEN
CLOUDFLARE_AI_TOKEN + CLOUDFLARE_ACCOUNT_ID
HUGGINGFACE_TOKEN
MISTRAL_API_KEY
SAMBANOVA_API_KEY
OPENROUTER_API_KEY
ZAI_API_KEY
SILICONFLOW_API_KEY
AION_LABS_KEY
OLLAMA_CLOUD_KEY
MODELSCOPE_TOKEN
KILO_CODE_TOKEN

Set whichever ones you have — providers without a key are skipped in the fallback chain.

Contributing

Code lives here. Provider / model / rate-limit data lives in chirag127/oriz-ai-providers-data (CC0-licensed).

• To add a provider: open a PR on the data repo with a new entry in providers.json + models.json + env-vars.json + (optionally) priority.json. Then open a PR here with a new src/providers/<slug>.ts adapter (usually 2 lines via the factory).
• To fix a rate limit: PR the data repo only — no code release needed.

License

MIT. See LICENSE.