2.2.3

Changes

• Bug-fix: Fix cost calculation coming from litellm
• Bug-fix: Safeguarding application workflow
• Bug-fix: exclude vital columns while masking like model_name, cost, latency, span_id, trace_id etc.
• Feat: set model_cost as no op function
• Bug-fix: export all columns without any filter
• Bug-fix: fix total cost value in the trace details

2.2.1

Changes

• Feat: Unify the trace format for RAG, Agentic Traces
• Feat: Add feature to automatically refresh token after every 6 hrs
• Feat: Add greater support to capture errors
• Bug: Fix for CSV upload of numerical, categorical values
• Bug: Fix for metric execution error with "_" in column names
• Bug: Fix external_id inconsistencies
• Bug: Fix Add proper span hash ids

Full Changelog: v2.1.7.4...v2.2.1

2.1.7.4

Changes:

• add_metadata
• mask traces
• support for error capturing for RAG
• Improve fallback for token counting

Full Changelog: v2.1.7.1...v2.1.7.4

2.1.7.1

Changes

• Feat: Support adding external_id
• Feat: Add post-processing hook, PII removal hook
• Feat: Trace Upload Consistency on Load
• Feat: RAG-Tracing using OpenInference
• Feat: Test cases, CI/CD. Pipeline
• Bug-fix: list_dataset() to work for large number of datasets
• Bug-fix: Indexing Error in Agentic Tracing
• Bug-fix: Check for crashed when defining tracer without metadata key.
• Bug-fix: add_context not working for langchain rag

Full Changelog: 2.1.6.4...v2.1.7.1

2.1.6.4

What's Changed

1. fix: Corrected total_cost and total_token calculation in custom agentic traces

   • Previously, these values were being incorrectly calculated or displayed.
   • Now, the logic ensures accurate display of total cost and token usage.

2. feat: Associate model with response and add model_name metadata

   • Associated the LLM model used with its corresponding response to align with backend changes and provide a more complete data structure.
   • Introduced a new metadata field, model_name, specifically for analysis purposes. This will allow for easier tracking and reporting of model usage.

3. fix: Graceful handling of missing LLM response data

   • Improved error handling to prevent crashes when model cost, token, or latency cannot be identified from the LLM response.
   • The system now gracefully handles cases where any of these values are missing, ensuring stability and continued operation.

2.1.6.3

What's Changed

• Bump litellm from 1.42.12 to 1.61.15 by @dependabot in #193
• Bump langchain-core from 0.2.11 to 0.2.43 by @dependabot in #194
• Make timeout configurable for agentic/<framework>, Add support for set custom model cost for langchain RAG by @kiranscaria in #204

Full Changelog: 2.1.6.2...2.1.6.3

2.1.6.2

Changes

• Add support for tracing OpenAI Agents SDK
• Update trace schema to:
  • moved recorded_on to schema_type timestamp
  • add total_cost, total_tokens as numerical metadata
  • add model_name as categorical metadata
• Bug-fix in input_guardrails related to trace_id is None

2.1.6

What's Changed

• Add auto-instrumentation support for:
  • Langgraph
  • Langchain
  • CrewAI
  • Haystack
  • SmolAgents
• Add support for workflow (data collection) for auto-instrumentation
• Improved the guardrails flow
• Relaxed the dependencies, removed stale dependencies
• Add examples for multiple agentic frameworks
• Multiple bug-fixes

v2.1.5

Changes:

• Improve synthetic data generation
• Update redteaming
• Multiple bug-fixes
• Improve support for llamaindex tracing

v2.1.4

What's Changed

• support for trace_custom
• add workflow component
• add support for azure-openai for llm tracing
• bug-fix: resolve issues for cost, token
• made metadata & pipeline optional in trace definition
• support for dynamic update of dataset name after initialisation
• support to add_metrics locally
• bug-fix: resolve issues in code zip causing same code hash id
• bug-fix: resolve duplicate metrics added
• add support to debug using DEBUG=1
• add support to save code from colab & jupyter notebook.
• add support to trace file input, output interactions
• add support to concatenate _{index} to duplicate metric names in span
• unified time format across all files

New Contributors

• @keetrap made their first contribution in #136
• @joelrobin18 made their first contribution in #105
• @ujjman made their first contribution in #102
• @muditgaur-1009 made their first contribution in #111
• @NastyRunner13 made their first contribution in #114
• @VijayRagaAI made their first contribution in #143
• @Ritika1311 made their first contribution in #150