Integrations
Anthropic
Track the anthropic SDK (Claude). Coming after beta.
Not covered by beta support
The beta launch covers Gemini, LangChain and LangGraph. The Anthropic integration exists and works for the cases below, but it is not yet part of what we test and support for beta. Expect changes.
Install
pip install impraint-guard anthropicPatch or wrap
import anthropic
from impraint_guard import Guard
from impraint_guard.integrations.anthropic import patch_anthropic
guard = Guard(agent_id="claude-agent", api_key="...")
# Option 1: process-wide
patch_anthropic(guard)
client = anthropic.Anthropic()
# Option 2: a single client
client = guard.wrap_anthropic(anthropic.Anthropic())Both messages.create(...) (streaming or not) and the messages.stream(...) context manager are tracked.
Known gaps
- Behaviour when Claude is called through LangChain's
ChatAnthropicwhile the raw patch is active has not been verified yet. - Extended-thinking token accounting has not been verified yet.
- Claude on cloud gateways (Vertex AI, Bedrock, Azure) is not yet supported.
- Prompt checks (
on_input) exist onguard.wrap_anthropic, but they are not tested or supported in beta.patch_anthropichas no prompt checks.
See Compatibility for the full matrix.