Skip to content

Spending limits for Anthropic calls in Python

Anthropic Messages, sync and async, non-streaming and streaming. Verified locally with synthetic provider transports and real local KazenAI services, and also verified against the live OpenAI and Anthropic APIs on 2026-10-12 (gpt-6-luna, claude-sonnet-5-5).

from kazenai_finops import monitor, BudgetExceeded
import anthropic
client = monitor(anthropic.Anthropic(max_retries=0), agent_id="support", max_budget_usd=1.0)
# Non-streaming
client.messages.create(
model="claude-haiku-4-5",
max_tokens=256,
messages=[{"role": "user", "content": "Hello"}],
)
# Streaming via create(stream=True)
for event in client.messages.create(
model="claude-haiku-4-5",
max_tokens=256,
stream=True,
messages=[{"role": "user", "content": "Hello"}],
):
pass
# Streaming via messages.stream context manager
with client.messages.stream(
model="claude-haiku-4-5",
max_tokens=256,
messages=[{"role": "user", "content": "Hello"}],
) as stream:
for event in stream:
pass
final = stream.get_final_message()

Early context exit or close() leaves financial reconciliation pending when final usage is unknown.

KazenAI merges input usage from message_start with later output usage from message_delta or the final Message. Text, tool JSON, thinking, and signature deltas contribute to the optional local cutoff estimate; provider usage remains the authority for settlement.

Wrap anthropic.AsyncAnthropic the same way and await the calls:

import asyncio
import anthropic
from kazenai_finops import monitor, BudgetExceeded
async def main() -> None:
client = monitor(anthropic.AsyncAnthropic(max_retries=0), agent_id="support", max_budget_usd=1.0)
try:
msg = await client.messages.create(
model="claude-haiku-4-5",
max_tokens=256,
messages=[{"role": "user", "content": "Hello"}],
)
stream = await client.messages.create(
model="claude-haiku-4-5",
max_tokens=256,
stream=True,
messages=[{"role": "user", "content": "Hello"}],
)
async for event in stream:
pass
except BudgetExceeded as exc:
print(f"Denied before dispatch: {exc}")
finally:
await client.close()
asyncio.run(main())

Both calls send the same prompt. That’s fine: by default the loop guard records a repeat and sends it. With monitor(..., loop_guard="block") the second call raises LoopDetected (from kazenai import LoopDetected) before dispatch, so catch it next to BudgetExceeded. See Repeated requests.

Surface Status
Bedrock / Vertex Anthropic wrappers UNSUPPORTED

See the full matrix: Supported surfaces.