API

Walk a workflow against the live API, or browse the contract. What you see is what's running on ….

More workflows

Each is a runnable, multi-step path through the API.

Use your AI assistant against this API

Preview

Point Claude (or any tool-using LLM) at your airbrx PAT + the OpenAPI spec and ask in plain English — "which tenant has the most cache misses?", "draft a cache rule for my dim_orders table." A starter snippet and system prompt are below.

Starter snippet — Python + Anthropic SDK

# Stub — fill in with a working example before shipping.
import anthropic
import requests

AIRBRX_BASE  = "https://api.airbrx.ai"
AIRBRX_TOKEN = "pat_..."  # create one at /api/pats.html

def airbrx_get(path):
    r = requests.get(f"{AIRBRX_BASE}{path}",
                     headers={"Authorization": f"Bearer {AIRBRX_TOKEN}"})
    r.raise_for_status()
    return r.json()

client = anthropic.Anthropic()

resp = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system=open("system_prompt.md").read(),
    tools=[{
        "name": "airbrx_get",
        "description": "GET an airbrx admin API path; returns JSON.",
        "input_schema": {
            "type": "object",
            "properties": {"path": {"type": "string"}},
            "required": ["path"],
        },
    }],
    messages=[{"role": "user",
               "content": "Which tenant has the most cache misses this year?"}],
)
print(resp.content)

Starter system prompt

You are an analyst working against the airbrx admin API.

Auth is a bearer PAT, exposed through the airbrx_get tool. Useful endpoints:
  GET /config/tenants
  GET /tenants/{id}/summaries/{year}
  GET /tenants/{id}/summaries/{year}/opportunities
  GET /config/tenants/{id}/rules

When the user asks about caching effectiveness, savings, or "where can we
cache more," fetch the year-summary and opportunities for the relevant
tenant(s). When asked to draft a rule, propose JSON that conforms to the
tenant rules schema and explain which queries it would match.

Answer in plain English. Cite the endpoints you called.

This is a placeholder card — content + a runnable Node version land in a follow-up.