AI spend, governed.Not just observed.
Spillway is the control plane between your apps and every model — budgets that block, approvals that gate, and routing that saves.
Every model. One line of code.
The over-budget request is refused before the spend. It costs exactly $0.
budgets cascade org → team → key · hard-stop, alert, or monitor · fail-open by default, fail-closed per key
See every dollar. Govern the next one.
Live spend by org, team, and key — every request priced from a unit-price snapshot, and the monthly statement reconciles to the cent.
Explore the consoleReplay any decision.
Every prompt, its tier, the model it took, and what you saved.
See the route logBudgets that block. Approvals that gate.
Limits cascade org → team → key and hard-stop at the gateway. Big spends wait for a human yes.
See governanceSpend that needs a human gets one.
The spike is caught mid-flow.
The statement finance can use.
“Summarize this 40-page vendor contract and flag every clause that creates liability for us.”
Where it sits.
A base-URL swap puts the gateway in the path. Everything else follows from that.
Change one line.
Point your OpenAI client at Spillway. Keep everything else.
Read the docs# pip install openai — no new SDK from openai import OpenAI client = OpenAI( base_url="https://api.spillway.dev/v1", api_key=os.environ["SPILLWAY_KEY"], ) resp = client.chat.completions.create( model="auto", # let Spillway choose messages=messages, extra_body={"policy": "balanced"}, ) # resp.spillway → {tier, model, saved_usd} print(resp.spillway.model) # gpt-4.1-mini
Free to start.
Governance when you need it. Never a percent of your spend.
Enterprise — in-region, on-prem, SLA · Talk to us