robusta-dev/holmesgpt 0.14.1

on GitHub

🚀 Release Notes

This release brings four major improvements (and many small improvements):

1. A much improved Tempo integration
2. Many optimizations to context window handling - these are especially impactful when using HolmesGPT with metrics and trace data. They dramatically increase the accuracy of Holmes' answers
3. Improvements to the DataDog integration - we made large number of small changes to improve overall accuracy.
4. Support for Sonnet 1M token model - we are looking for Robusta SaaS who would like to test it

🌟 Highlight - Improved Tempo Integration

You can now ask HolmesGPT questions like:

• "Why was there an increase in latency?"
• "Find the cause of recent HTTP 500 errors."

Please share your feedback on the Tempo integration! We're going to focus more on traces in the next few releases.

What's Changed

• [ROB-2024] fix assume_role eks by @Avi-Robusta in #936
• fix datadog docs and remove experimental tag by @aantn in #939
• tracing tool call result and errors correctly by @Sheeproid in #937
• bump prometrix version by @Avi-Robusta in #940
• Don't fail the evals workflow on regressions by @aantn in #944
• feat: Support llm-based message summarization by introducing Transformer mechanism by @nilo19 in #695
• chore(azure openai): support api_base and api_version in config by @mainred in #881
• don't trigger build on push non-master branch by @mainred in #938
• chore: remove packages not used in holmesgpt by @mainred in #931
• add free form tempo toolset by @Sheeproid in #948
• Load default robusta model from API by @moshemorad in #946
• Docs: Add instructions on choosing model when using HTTP API by @aantn in #891
• updated docs for new tools and configuration by @Sheeproid in #951
• [ROB-1896] fixes edge case prometheus alert start time by @Avi-Robusta in #947
• [ROB-1738] Truncated metadata support by @Avi-Robusta in #841
• Implement LLM Model Registry by @moshemorad in #954
• add docs on robusta ai by @aantn in #943
• Better ollama support by @aantn in #878
• ROB-2005: improve error handling for streaming (slackbot) by @nherment in #935
• ROB-2116: prevent tool calls responses that are too big by @nherment in #956
• Improve prometheus truncation by @aantn in #955
• Sent truncation data to sentry by @moshemorad in #958
• fix: rename long file name to abode by windows limit by @nherment in #959
• chore: reduce max prometheus query chars to 50k by @nherment in #960
• ensure token_count is added to the final message by @nherment in #963
• Small eval improvements + investigation-refusal eval by @aantn in #966
• Fix bug causing massive evals slowdown by @aantn in #968
• Add max tokens to holmes response by @moshemorad in #970
• Add max output tokens to holmes response by @moshemorad in #972
• Respect DEFAULT_LOG_LIMIT in Loki + fix Datadog eval by @aantn in #971
• Improve Prometheus integration by @aantn in #967
• Update ai provider docs by @pavangudiwada in #952
• improve prometheus integration by @aantn in #975
• Minor fixes - mostly to cli output by @aantn in #976
• ROB-2117 allow override context size with model list by @RoiGlinik in #974
• Datadog improvements by @aantn in #973

Full Changelog: 0.13.3...0.14.1