Quick start guide

Install the wrapper

pip install tokeven

Wrap your client

from anthropic import Anthropic
from tokeven import TokevenAnthropic

client = Anthropic()
tracked = TokevenAnthropic(
    client,
    ingest_token="tkv_ing_...",
    base_url="https://api.tokeven.com",
)

Leave base_url out and the SDK falls back to http://localhost:8000, where nothing is listening. The emitter never raises on a failed push (instrumentation must not break the call it wraps), so events spool to disk and the dashboard stays empty with no error anywhere.

Make a call

Use tracked exactly like your normal client. Cost, tokens, model, and latency are captured automatically.

response = tracked.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)

Your prompt and completion content never reaches Tokeven - only metadata (token counts, model ID, latency, cost) is sent. (Two opt-in features - prompt revision and advisor.improve() - can send prompt text to Anthropic under your own API key; nothing prompt-related ever goes to Tokeven.)

View your data

Log in to the dashboard to see your first tracked call. Data appears within seconds.