TruLens 2.7.1

Bug Fixes

• Fix papercuts with Metric API by @sfc-gh-jreini in #2362

Tech Debt Reduction

• refactor "v2" feedback organization from monorepo by @sfc-gh-jreini in #2354

Test Fixes

• Fix race condition for otel test db creation by @sfc-gh-jreini in #2368

Docs and Examples

• update experimental examples and test with metricconfig by @sfc-gh-jreini in #2353

Full Changelog: trulens-2.7.0...trulens-2.7.1

TruLens 2.7.0

New Unified Metric API

We're introducing the Metric class as the new primary API for evaluation metrics in TruLens. This unified interface replaces both Feedback and MetricConfig with a cleaner, more explicit design.

Why This Matters

• Clearer, more readable code: Replace chained .on_*() methods with an explicit selectors={} dictionary—no more guessing which argument maps to which selector
• Single unified class: No need to choose between Feedback and MetricConfig for different use cases
• Better OTel integration: Native support for selecting data from OpenTelemetry spans

As a bonus, the same Metric code is now portable to Snowflake environments without modification.

Quick Example

Before (Feedback):

f_groundedness = (
    Feedback(provider.groundedness_measure_with_cot_reasons, name="Groundedness")
    .on_context(collect_list=True)
    .on_output()
)

After (Metric):

f_groundedness = Metric(
    implementation=provider.groundedness_measure_with_cot_reasons,
    name="Groundedness",
    selectors={
        "source": Selector.select_context(collect_list=True),
        "statement": Selector.select_record_output(),
    },
)

The selectors dict makes it explicit which function parameter receives which data.

Migrating from MetricConfig

Before (MetricConfig):

from trulens.core.feedback.custom_metric import MetricConfig

config = MetricConfig(
    metric_name="word_count",
    metric_implementation=lambda text: len(text.split()),
    selectors={"text": Selector.select_record_output()},
)
feedback = config.create_feedback_definition()

After (Metric):

from trulens.core import Metric, Selector

metric = Metric(
    name="word_count",
    implementation=lambda text: len(text.split()),
    selectors={"text": Selector.select_record_output()},
)

Backward Compatibility

The existing Feedback class continues to work with a deprecation warning, giving you time to migrate:

# Still works, but deprecated
from trulens.core import Feedback
f = Feedback(provider.relevance).on_input().on_output()

See the migration guide for detailed instructions on updating your code.

Bug Fixes

• fix: correct typo 'occurences' to 'occurrences' by @thecaptain789 in #2349
• Drop pkg_resources dependency from trulens-core by @debu-sinha in #2351
• fix: add circular-reference-safe repr to SerialModel (#1862) by @sfc-gh-jreini in #2356
• Forward litellm routing params (api_base, api_key, etc.) to completion calls by @sfc-gh-jreini in #2357
• fix: guard against None model_name in OpenAI endpoint (#1700) by @sfc-gh-jreini in #2358

Docs

• New Example and Docs Updates for usage in MLFlow by @sfc-gh-jreini in #2359

New Contributors

• @thecaptain789 made their first contribution in #2349

Full Changelog: trulens-2.6.0...trulens-2.7.0

TruLens 2.6.0

What's Changed

Agent Skills and Agents.MD

• agents.md and skills by @sfc-gh-jreini in #2337
• Skill and helpers for more beautiful tracing by @sfc-gh-jreini in #2340

Support for waiting for evaluation completion

• Fix bug and allow waiting in otel by @sfc-gh-jreini in #2339

Postgres Support

• Postgres support by @sfc-gh-jreini in #2346

Docs

• Update the docs to make it easier to reflect contribution process and current state of trulens by @sfc-gh-jreini in #2330
• [docs] improve crosslinking between attribution instrumentation and eval by @sfc-gh-jreini in #2336

Bug Fixes

• Fix user prompt relevance template in llm_provider.py address issue 2334 by @sfc-gh-jreini in #2341
• #2331 Update plotly to >= 6.0.0 by @jacobus-herman in #2332
• fix safe_getattr to handle more types by @sfc-gh-jreini in #2338

Release blog: https://www.trulens.org/blog/2026/02/03/trulens-26-skills-for-ai-coding-assistants-postgresql-support-and-more/

TruLens 2.5.3

What Changed

• Direct registration of LangGraph graph compatibility with AI Observability Run by @sfc-gh-dhuang in #2311
• Additional Instructions for all feedback functions by @sfc-gh-ajia in #2313
• Loosen trace compression by @sfc-gh-jreini in #2312

Bug Fixes

• [Minor fix] Remove staticmethod descriptor for trace_with_run by @sfc-gh-dhuang in #2319
• Fix mcp span type by @sfc-gh-jreini in #2324
• Keep host during verification. This also will have downstream effects in that it will use that to build the SQLAlchemy URL. by @sfc-gh-dkurokawa in #2323
• Fix: Skip instrumentation of non-callable objects (e.g., CallTypes enum) by @debu-sinha in #2328

Docs

• Blog: Celebrating 3,000 Stars by @sfc-gh-jreini in #2321

TruLens 2.5.2

What's Changed

• fix: add additionalProperties: false to structured output schemas for Databricks compatibility by @debu-sinha in #2308
• Make command for updating meta.yaml files by @sfc-gh-jreini in #2309

New Contributors

• @debu-sinha made their first contribution in #2308

Full Changelog: trulens-2.5.1...trulens-2.5.2

TruLens 2.5.1

TruLens 2.5.1 Patch Release

Bug Fixes

• [Snowflake] Enable get_records to fetch serverside metrics. by @sfc-gh-dkurokawa in #2293
• [Snowflake] Enable serverside metric computation kickoff from .py scripts by @sfc-gh-dhuang in #2297
• Fix internal method to comply with pydantic change introduced in pydantic 2.10 by @sfc-gh-jreini in #2296

TruLens v2.5.0

What's Changed

• Langchain 1.0 support by @sfc-gh-jreini in #2279
• Update pip install in notebook command with version constraints for langchain packages by @sfc-gh-jreini in #2275
• Remove noisy log msg by @sfc-gh-dhuang in #2277
• track eval costs, surface total costs on leaderboard by @sfc-gh-jreini in #2278

New examples

• Example of agent human annotation flow by @sfc-gh-jreini in #2237
• Updated langchain quickstart with newer agentic rag langchain flow for langchain 1.0 by @sfc-gh-jreini in #2275
• Example of langchain astream by @sfc-gh-jreini in #2275

Bug Fixes

• Usage and cost metrics for Google Gemini model generations by @sfc-gh-ajia in #2272
• Add force_flush() at the end of client-side custom metric computation by @sfc-gh-dhuang in #2289
• Resolve custom_instructions error for non-GPA feedback functions by @sfc-gh-ajia in #2284
• [SNOW-2655386] [AI Observability] - Context Relevance and Groundedness Not Computing for Virtual App by @sfc-gh-dhuang in #2288
• Docs fixups by @sfc-gh-jreini in #2280

TruLens v2.4.2

What's Changed

• Tracing for LlamaIndex AgentWorkflow + notebook example with GPA evals by @sfc-gh-jreini in #2249
• Tracing and Eval for MCP in LangGraph by @sfc-gh-jreini in #2257
• Updating GPA prompts by @sfc-gh-ajia in #2261
• quiet logger on context manager warning by @sfc-gh-jreini in #2262
• [Snowflake] SDK changes to use latest field mode in RunConfig by @sfc-gh-dhuang in #2254

Bug Fixes

• Fix langchain compatibility for langchain 1.0 by @sfc-gh-jreini in #2267
• Fix join logic and app_version reference in _check_cross_format_records by @sfc-gh-jreini in #2266
• Fix conda build issue by limiting build to python < 3.13 for now. by @sfc-gh-dkurokawa in #2264
• Fix bugs with runs and metric computation by @sfc-gh-jreini in #2263
• Fix snowflake connector SQL util (to_pandas issue) by @sfc-gh-dhuang in #2265

Full Changelog: trulens-2.4.1...trulens-2.4.2

Trulens v2.4.1

What's Changed

• [Snowflake] web search agent example with snowflake ai obs by @sfc-gh-jreini in #2232
• Handle timestamps in get_events for Snowflake AI observability event table. by @sfc-gh-dkurokawa in #2226
• data agent with cortex example by @sfc-gh-jreini in #2236
• Update meta.yaml files for release 2.4.0 by @sfc-gh-dhuang in #2238
• OpenAI cost tracking w/ the new Responses API by @sfc-gh-dhuang in #2239
• Fix Cortex cost/usage tracking with latest REST API by @sfc-gh-dhuang in #2240
• [Feat] Update record viewer default expanded depth to 2 levels by @sfc-gh-nvytla in #2244
• [Snowflake] Virtual Run (batch eval run on data in existing table in Snowflake) by @sfc-gh-dhuang in #2224
• Fix get_records_and_feedbacks for AI Observability by @sfc-gh-dhuang in #2245
• [Snowflake] Add get_records and get_record_details APIs to AI Observability Run by @sfc-gh-dhuang in #2246
• Denoise trulens logs by @sfc-gh-dhuang in #2243
• Get Streamlit UI Record Viewer to work with AI Observability. by @sfc-gh-dkurokawa in #2233
• Create an "SPCS" mode for the dashboard that allows authentication through an OAuth token mounted into the docker container. by @sfc-gh-dkurokawa in #2241
• Add run_name to get_records_and_feedback. by @sfc-gh-dkurokawa in #2247
• Convert execute_query to return pd.DataFrame instead of List[Row]. by @sfc-gh-dkurokawa in #2251
• Refresh Snowflake connection when closed. by @sfc-gh-dkurokawa in #2250
• Fix - new Run API should take run_name when filtering records by @sfc-gh-dhuang in #2252
• [Feat] Add TS/TC/TQ judges by @sfc-gh-nvytla in #2256
• Avoid problematic transformers version requirement as there's an issue with its grouped optional dependencies. by @sfc-gh-dkurokawa in #2255

Full Changelog: trulens-2.4.0...trulens-2.4.1

trulens-2.4.0

New Features

• Tracing for LlamaIndex Workflows + Improved Async OpenAI support by @sfc-gh-jreini in #2212

• Add TruSession::compute_feedbacks_on_events function to compute feedbacks on arbitrary records and feedbacks. by @sfc-gh-dkurokawa in #2206

• trace compression by @sfc-gh-jreini in #2225

• Reranker span type type and inclusion of reranker span type in LlamaIndex auto-instrumentation @sfc-gh-jreini in #2222

• [Snowflake] Have AI observability work with get_records_and_feedback. by @sfc-gh-dkurokawa in #2223

• [Snowflake] Implementing client side custom metrics for Snowflake batch evaluation flow by @sfc-gh-dhuang in #2201

• [Snowflake] Update model costs in cortex_model_costs.json by @sfc-gh-jreini in #2228

Examples

• web search agent evals by @sfc-gh-jreini in #2209
• Update additional notebooks for OpenTelemetry by @sfc-gh-jreini in #2222

Bug Fixes

• Fix json structured output for cortex complete by @sfc-gh-dhuang in #2214 and #2217
• Use Feedback class instead of underlying function when running evals by @sfc-gh-jreini in #2219
• Restrict streamlit-aggrid to < 1.1.8 to fix. by @sfc-gh-dkurokawa in #2227

Full Changelog: trulens-2.3.1...trulens-2.4.0