Putting a dollar cap on one Strands agent run

[domondi1]

A Strands agent decides for itself how many model calls a task takes, so a tool that keeps failing can turn one request into dozens of calls. Here's one way to put a hard dollar ceiling on a single run without a plugin or a change to the agent.

OpenAIModel passes client_args to the OpenAI client, so the run's id and budget go in default_headers, pointed at a small gateway running in the same process:

import inferrail
base_url = inferrail.start()  # gateway on a background thread, no second process

from strands import Agent
from strands.models.openai import OpenAIModel

def agent_for_run(run_id: str, budget_usd: str) -> Agent:
    model = OpenAIModel(
        client_args={"base_url": base_url, "api_key": "unused",
                     "default_headers": {"X-Inferrail-Attribute-Work-Id": run_id,
                                         "X-Inferrail-Budget-Usd": budget_usd}},
        model_id="gpt-4o-mini", params={"max_tokens": 200})
    return Agent(model=model, tools=[...])

Each call reserves its worst-case cost before it goes out, so it's checked before the money is spent rather than after. In a test with a tool that always answered "unavailable, try again", a $0.002 budget stopped the loop after 17 calls: the next call got a 402 before the provider and the event loop ended with EventLoopException. inferrail work <run id> shows what the run cost.

Caveats: Chat Completions only, set max_tokens, and only calls through that model count. Tested with strands-agents 1.57.1 and 1.57.2.

Full setup: guide. I maintain Inferrail (open source, Apache-2.0), so take the recommendation with that in mind. Curious how others bound the cost of a single run here.