llmkit-sdk 0.1.4

AI API gateway SDK with cost tracking, budget enforcement, and multi-provider routing for LLM applications

llmkit

Python SDK for LLMKit, the AI API gateway with cost tracking and budget enforcement.

Install

pip install llmkit-sdk

Local cost tracking (no proxy)

Drop-in cost tracking for any OpenAI-compatible SDK. No account, no proxy, no config:

from llmkit import tracked
from openai import OpenAI

client = OpenAI(http_client=tracked())
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)
# costs estimated locally from bundled pricing table

Collect costs with a callback:

costs = []
client = OpenAI(http_client=tracked(on_cost=costs.append))
# ... make requests ...
total = sum(c.total_cost for c in costs if c.total_cost)

Or estimate from any existing response:

from llmkit import estimate_cost

cost = estimate_cost(response)
print(f"~${cost.total_cost:.6f}")

Works with OpenAI, Anthropic, Groq, Together, Cohere, and Mistral SDKs.

Quick start (proxy mode)

from llmkit import LLMKit

client = LLMKit(api_key="llmk_...")
completion, cost = client.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)

print(completion.choices[0].message.content)
print(f"Cost: ${cost.total_cost:.4f} via {cost.provider}")

Session tracking

Group costs by agent run using sessions:

agent = client.session()

for task in tasks:
    completion, cost = agent.chat(
        model="gpt-4o",
        messages=[{"role": "user", "content": task}],
    )

print(f"Session total: ${agent.stats.total_cost:.4f}")
print(f"Requests: {agent.stats.request_count}")

Bring your own provider key

client = LLMKit(
    api_key="llmk_...",
    provider_key="sk-...",
)

Streaming with cost tracking

stream = client.chat_stream(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

print(f"\nCost: ${stream.cost.total_cost:.6f}")

Cost is captured from the final stream chunk's usage data. Cumulative totals are available on client.stats.

Escape hatch

The underlying OpenAI client is always accessible for anything the SDK doesn't cover:

stream = client.openai.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
    stream=True,
)

for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Async

from llmkit import AsyncLLMKit

client = AsyncLLMKit(api_key="llmk_...")
completion, cost = await client.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)

No SDK needed

LLMKit works with any OpenAI-compatible client. No pip install required:

from openai import OpenAI

client = OpenAI(
    base_url="https://llmkit-proxy.smigolsmigol.workers.dev/v1",
    api_key="llmk_...",
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)

License

MIT