Python SDK

Installation

pip install agent-cost-optimizer

Agentrics(api_key)

Create a single Agentrics instance at startup and use it to wrap all your AI clients.

from agent_cost_optimizer import Agentrics

agentrics = Agentrics(api_key="agntcs_sk_...")

wrap_anthropic(client, agent_id, agent_name?)

Wraps an anthropic.Anthropic() instance. The returned client has the same interface — use it exactly as you would the original.

from anthropic import Anthropic
from agent_cost_optimizer import Agentrics

agentrics = Agentrics(api_key="agntcs_sk_...")

client = agentrics.wrap_anthropic(
    Anthropic(),
    agent_id="support-classifier",
    agent_name="Support Classifier",   # optional, defaults to agent_id
)

# Use identically to the original client
response = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.content)

wrap_openai(client, agent_id, agent_name?)

Wraps an openai.OpenAI() instance. Intercepts all chat.completions.create() calls.

from openai import OpenAI
from agent_cost_optimizer import Agentrics

agentrics = Agentrics(api_key="agntcs_sk_...")

client = agentrics.wrap_openai(
    OpenAI(),
    agent_id="research-agent",
)

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Summarize this..."}],
)
print(response.choices[0].message.content)

Error handling

Tracking is fire-and-forget — if the Agentrics API is unreachable, the error is silently swallowed and your AI call is unaffected. If the underlying AI API raises an exception, the SDK records the failure (with success=False) before re-raising it.