AI Badgr

Drop-in replacement for OpenAI & Claude APIs. Lower cost. Lower latency. No code changes.

Just swap one URL:

import openai

client = openai.OpenAI(
    api_key="your-aibadgr-key",
    base_url="https://aibadgr.com/v1",  # ← this
)

That's it. Everything else stays the same.

---

Why AI Badgr?

LLM API bills compound fast. At scale, 429s and retry storms make it worse. AI Badgr runs on community GPU workers — giving you a cheaper, faster endpoint that's compatible with every OpenAI/Claude client you're already using.

Real numbers, not estimates. Upload your OpenAI or Anthropic usage export at aibadgr.com/bill-calculator and see your exact savings replayed against your real workload.

---

Quick Start

1. Get your API key → aibadgr.com/signup

2. Swap the base URL

Provider	Old	New
OpenAI	api.openai.com	aibadgr.com
Claude	api.anthropic.com	aibadgr.com

3. Verify your savings in 60 seconds

export OPENAI_API_KEY="sk-..."
export BADGR_API_KEY="your-aibadgr-key"
python examples/badgr-verify-demo.sh

Output:

OpenAI:  ttfb=410ms  total=2.9s  in=312  out=128  est_cost=$0.0009
Badgr:   ttfb=180ms  total=2.1s  in=312  out=128  est_cost=$0.0004
Diff:    -230ms ttfb, -0.8s total, -56% cost

---

Works with everything you're already using

Library	Works?
OpenAI SDK (Python + JS)	✅
Anthropic SDK	✅
LangChain	✅
LlamaIndex	✅
LiteLLM	✅
Vercel AI SDK	✅
Haystack	✅
AutoGen / CrewAI	✅
Dify, RAGFlow, Open WebUI	✅

See docs/examples.md for copy-paste snippets.

---

Pricing

Per-token. No subscriptions. No minimums.

Model	Input / 1M	Output / 1M
Llama 3.1 8B	$0.20	$0.20
Mistral Nemo	$0.30	$0.60
Llama 3.3 70B	$0.65	$1.20

---

Migrate your whole codebase in one line

curl -s https://aibadgr.com/tools/migrate-to-aibadgr.py | python - ./src

Preview first with --dry-run:

curl -O https://aibadgr.com/tools/migrate-to-aibadgr.py
python migrate-to-aibadgr.py ./src --dry-run

---

Earn from your idle GPU

Run the worker on any CUDA-capable machine. While it's idle, AI Badgr automatically lists it on Vast.ai, RunPod, Thunder Compute, and 7 other marketplaces. When a job arrives, it instantly unlists. Zero configuration.

docker run -d --name aibadgr-worker --restart unless-stopped --gpus all \
  -e API_BASE_URL="https://aibadgr.com" \
  michaelmanleyx/aibadgr-worker:latest

The worker auto-detects GPU memory and picks the best model tier.

---

Docs

• Getting Started
• Concepts — architecture, tiers, encryption, incident radar
• Examples — LangChain, Anthropic, TypeScript, RAG

Topics

llm rag ai openai anthropic gpu inference langchain litellm retry

---

Contributing

Found a bug, want an integration, or hitting 429s? Open an issue — the retry issue template is specifically for rate-limit pain. PRs welcome. See CONTRIBUTING.md.

License

MIT