feat(integrations): Add Anthropic Integration #2831
Conversation
This commit adds a basic integration for the anthropic sdk.
|
Hey @czyber ! Thanks for this great PR! This makes it then easier to review the integration and speeds things up so we can get this merged soon! |
|
Hey again @czyber , I made some changes to better deal with the spans. Everything looks great! One question: Do the test actual send data to Anthropic? Or does the "z" api_key enable some sort of test mode? (because we do not want to access the internet in our tests, if it is not absolutely necessary) |
|
Hey @antonpirker, Thank you for updating the PR! The tests do not send data to Anthropic, yet initializing the client with |
There was a problem hiding this comment.
This looks good! Thanks again for the contribution!
I will hold on merging this for a bit, because we released 2.0 yesterday and maybe we need to do some hot fixes.
We also have a Langchain integration in the works right now, and I would like to release both in the same version, so we have a kind of AI update :-)
sentry_sdk/integrations/anthropic.py
Outdated
| if input_tokens != 0: | ||
| span.set_data(PROMPT_TOKENS_USED, input_tokens) | ||
| if output_tokens != 0: | ||
| span.set_data(COMPLETION_TOKENS_USED, output_tokens) |
There was a problem hiding this comment.
Once langchain is merged, this should use the new helper function https://github.com/getsentry/sentry-python/pull/2911/files#diff-c566959db6701d8426b02bcf30caa0222d01aa9931ce9913b1a35fedfd0953c1R59
sentry_sdk/integrations/anthropic.py
Outdated
|
|
||
| with capture_internal_exceptions(): | ||
| span.set_data("ai.model_id", model) | ||
| span.set_data("ai.streaming", False) |
There was a problem hiding this comment.
It'd be nice to pull the tags we have in langchain too if any exist in the params: https://github.com/getsentry/sentry-python/pull/2911/files#diff-bba846cf7b623cb617bd628d482cd71fe41160f0cea1f48f321fcf5166472925R52-R62
sentry_sdk/integrations/anthropic.py
Outdated
| "ai.responses", | ||
| [{"type": "text", "text": complete_message}], | ||
| ) | ||
| span.set_data(TOTAL_TOKENS_USED, input_tokens + output_tokens) |
There was a problem hiding this comment.
New helper function again (after langchain merged)
| assert "ai.input_messages" not in span["data"] | ||
| assert "ai.responses" not in span["data"] | ||
|
|
||
| assert span["data"][PROMPT_TOKENS_USED] == 10 |
There was a problem hiding this comment.
Some of this has changed a bit in the langchain merge as well, it's span["measurements"] I believe
|
Hey @colin-sentry , I have the langchain PR on my mind, and will incorporate the requested changes as soon as it is merged 👍 |
|
fyi @czyber the langchain branch has been merged into master! |
This commit updates the anthropic integration to include the newest changes from the langchain integration
|
fyi @colin-sentry @antonpirker , I updated the PR to use the shared functionality for AI integrations that was introduced in the langchain integration PR. |
|
Yes! Great work everybody! Thanks a lot! |
This PR adds an anthropic integration. It supports the creation of messages in streaming and non-streaming mode.
It is heavily inspired/analogous to the OpenAI integration.
Docs PR: getsentry/sentry-docs#9469
General Notes
Thank you for contributing to
sentry-python!Please add tests to validate your changes, and lint your code using
tox -e linters.Running the test suite on your PR might require maintainer approval. Some tests (AWS Lambda) additionally require a maintainer to add a special label to run and will fail if the label is not present.
For maintainers
Sensitive test suites require maintainer review to ensure that tests do not compromise our secrets. This review must be repeated after any code revisions.
Before running sensitive test suites, please carefully check the PR. Then, apply the
Trigger: tests using secretslabel. The label will be removed after any code changes to enforce our policy requiring maintainers to review all code revisions before running sensitive tests.