Every example is a minimal, runnable recipe with its own README and expected output. Most need only two terminals: npx @axiorank/gateway in one, the example in the other.
| Example | What it shows |
|---|---|
| python-openai | The 4-line base_url swap with the official OpenAI Python SDK: one clean call, one blocked injection |
| node-openai | The same two beats with the official OpenAI Node SDK |
| vercel-ai-sdk | generateText through the gateway, plus @axiorank/ai-sdk middleware for in-process guardrails |
| langchain | ChatOpenAI through the gateway, plus the langchain-axiorank agent middleware |
| litellm | LiteLLM as the client SDK with the gateway as the governance layer, plus the axiorank guardrail for LiteLLM proxy |
| local-ollama | Fully local stack with docker compose: Ollama + guardrails + signed receipts, zero cloud, no API key |
| agent-frameworks | CrewAI and AG2 agents behind the gateway with one base_url change |
| receipts-in-ci | A GitHub Actions job that proves what its AI step did: verify the receipt chain, upload it as an artifact |
| canary-routing | Weighted round_robin as a model canary (95/5), watched via the x-axiorank-route-target header |
| coding-agents | Pointing Aider, Continue, and Cursor at the gateway, with honest notes on what does not work |
Missing a recipe? Open an issue or send a PR; the provider preset template is the easiest first contribution.