feat(node): Add Anthropic AI integration #17348
Conversation
RulaKhaled
force-pushed
the
anthropic-integration
branch
from
11ee820 to
f88aef1
Compare
size-limit report 📦
|
RulaKhaled
force-pushed
the
anthropic-integration
branch
from
59dda8c to
7feef93
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
![seer-by-sentry[bot]](https://avatars.githubusercontent.com/in/801464?s=60&v=4){kind=small}
RulaKhaled
force-pushed
the
anthropic-integration
branch
from
77ee440 to
fdb0059
Compare
| options?: AnthropicAiOptions, | ||
| ): (...args: T) => Promise<R> { | ||
| return async function instrumentedMethod(...args: T): Promise<R> { | ||
| const finalOptions = options || getOptionsFromIntegration(); |
There was a problem hiding this comment.
I would rename this a bit as it's not entirely clear from the function name what this function does. It's basically just checking if the options allow the inputs/outputs to be recorded.
| const finalOptions = options || getOptionsFromIntegration(); | |
| const finalOptions = options || getRecordingOptionsFromIntegration(); |
And could it be that this is the same function?
https://github.com/getsentry/sentry-javascript/pull/17348/files#diff-c9e6a17a3ac925c55873d82ff00c973c66d21433aec89430112dc3f10ff849cfR27
There was a problem hiding this comment.
I like the new name! And yah only one in core and one in node, not worth exporting as they are not an exact match
| response.usage.cache_creation_input_tokens, | ||
| response.usage.cache_read_input_tokens, | ||
| ); | ||
| } |
There was a problem hiding this comment.
Bug: Incorrect Token Count Due to Parameter Misuse
In addResponseAttributes, setTokenUsageAttributes receives response.usage.cache_read_input_tokens for its cachedOutputTokens parameter. Since cache_read_input_tokens are actually input tokens, this leads to an incorrect total token count. The cachedOutputTokens parameter name is also misleading.
Additional Locations (1)
| // Models.countTokens | ||
| if ('input_tokens' in response) { | ||
| span.setAttributes({ [GEN_AI_RESPONSE_TEXT_ATTRIBUTE]: JSON.stringify(response.input_tokens) }); | ||
| } |
There was a problem hiding this comment.
Bug: Token Count Attribute Misuse
The GEN_AI_RESPONSE_TEXT_ATTRIBUTE is incorrectly set with input_tokens for Models.countTokens responses. This attribute is meant for actual response text, not numerical token counts, which are already captured by setTokenUsageAttributes.
This PR adds official support for instrumenting Anthropic AI SDK calls in Node with Sentry tracing, following OpenTelemetry semantic conventions for Generative AI.
We instrument the following Anthropic AI SDK methods:
Supported in:
The anthropicAIIntegration() accepts the following options: