Dashboard Get started

Python SDK

Drop-in wrappers for Anthropic, OpenAI, xAI, Google, AWS Bedrock, Cohere, and Mistral that capture every call to your TokenID dashboard.

Installation

pip install --extra-index-url https://tokenid:@pypi.token.audit.id/simple/ tokenid

What gets captured

Field Description
provider anthropic, openai, xai, google, aws_bedrock, azure_openai, cohere, mistral
model Exact model string from the response
input_tokens Prompt tokens
output_tokens Completion tokens
cache_read Anthropic prompt cache read tokens
cache_creation Anthropic prompt cache creation tokens
reasoning_tok OpenAI reasoning tokens (o-series), Gemini thoughtsTokenCount
cached_tok OpenAI cached prompt token count
is_batch / batch_id Set automatically when the wrapper detects a Batch API response — cost is computed at 50% rate
team_id / feature Optional kwargs for sub-organization attribution (see below)
cost_usd Computed from live rate card at call time
session_id Your identifier — groups calls into sessions
tool_name / tool_input / tool_output Per tool_use and tool_result block
duration_ms Wall-clock latency
stop_reason end_turn, max_tokens, etc.

Wrappers

Anthropic

JournaledAnthropic

OpenAI

JournaledOpenAI

 xAI / Grok

JournaledGrok

 AWS Bedrock

JournaledBedrock

 Cohere

JournaledCohere

 Mistral

JournaledMistral

Team and feature labels

Pass optional team_id / feature kwargs to the wrapper constructor (applies to every call) or per-call (overrides the constructor value):

client = JournaledAnthropic(
    api_key="sk-ant-…",
    team_id="platform-ai",
    feature="rag-rerank",
)

Both columns are indexed and queryable on dashboard breakdown endpoints. Use them for internal chargeback or product-level cost attribution.

Streaming

All wrappers support streaming responses. Usage is captured from the final message_delta event.

with client.messages.stream(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Error handling

If the ingest call to TokenID fails, the original API call still succeeds and your code runs normally. TokenID never blocks your critical path.