mlflow

MLflow 3.12.0 includes several major features and improvements

Major New Features

• 🖼️ Multimodal Tracing: Users can now store multimodal content in tracing spans as artifact attachments instead of inline binary data. We've also patched the UI to support the new mlflow-attachment:// style URI, with rich rendering available for PDFs, audio, and images.
• 🤖 Codex, Gemini, Qwen coding agent tracing support: Similar to our Claude Code tracing integration, we've now added support for the Codex, Gemini, and Qwen coding agent platforms as well!
• 🛡️ Gateway guardrails: You can now set guardrails on your gateway endpoints to prevent unsafe or non-compliant model inputs and outputs. Try it out in the MLflow UI!
• ⚡ Trace table pagination: The traces tab is now paginated, rather than fetching all traces up to a limit of 1000. This improves initial load time, and makes the page feel more responsive overall.

Breaking Changes

• [Scoring] Deprecate enable_mlserver in pyfunc serving backend (#22994, @B-Step62)

Other Assorted Features & Improvements:

• [UI] Add coding agents section to AI Gateway quick start (#23006, @TomeHirata)
• [Tracing] feat: record caller in gateway traces from request headers (#22926, @TomeHirata)
• [] Run guardrails on passthrough endpoints; skip response_format for non-chat payloads (#22856, @TomeHirata)
• [] Gateway: preserve client auth header for subscription-based CLI tools (claude-cli, Codex-Desktop, GeminiCLI) (#22915, @TomeHirata)
• [Tracing] Expose Codex and Qwen Code hooks as installable CLI binaries (#22853, @kriscon-db)
• [Tracking / UI] Add Portkey as a supported AI Gateway provider (#22830, @sairavuri-sudo)
• [Tracking] Cache successful basic-auth credential checks to eliminate per-request PBKDF2 (#22817, @PattaraS)
• [] Add last_updated_at field to model catalog entries (#22838, @copilot-swe-agent)
• [Evaluation] Enable third-party scorer registration in OSS MLflow (#22634, @smoorjani)
• [] Add platform-side telemetry fields to Gateway invocation and budget events (#22557, @PattaraS)
• [Tracing] Auto-start SQL warehouse before V4/V5 MLflow tracing calls (#22798, @artjen)
• [Tracing] Migrate Claude Code tracing to TypeScript-based plugin for simpler installation and runtime handling. (#22338, @B-Step62)
• [Tracing] Add TypeScript Qwen Code tracing via Stop hook (#22411, @kriscon-db)
• [] Support model_kwargs in DeepEval scorers for LLM parameter control (#22494, @debu-sinha)
• [] Add Application Default Credentials auth mode for Vertex AI gateway (#22754, @harupy)
• [UI] Reorganize Settings into section-based routing with sub-sidebar navigation (#22743, @serena-ruan)
• [Tracing] Add TypeScript Codex CLI tracing via notify hook (#22410, @kriscon-db)
• [Tracing / UI] Add Share button to trace detail view (#22608, @alkispoly-db)
• [Docs / Models] Add mlflow.diffusers flavor for diffusion model LoRA adapters (#22253, @Rasaboun)
• [Tracing] Add size limit for trace attachments (#22575, @kriscon-db)
• [Tracing] Add GeminiCliTranslator for Gemini CLI OTLP span type mapping (#22409, @kriscon-db)
• [Tracing] Add JSON OTLP encoding support for trace ingestion (#22408, @kriscon-db)
• [Tracing] Extract service.name from OTLP resource attributes for usage telemetry (#22407, @kriscon-db)
• [Tracing] Add MLflow tracing spans to guardrail execution in gateway (#22581, @TomeHirata)
• [Tracing / UI] Add rendering size guards for large media content (#22574, @kriscon-db)
• [Tracking] #21037 Add presigned upload URL endpoint for S3 artifact uploads (#21039, @henishborad)
• [UI] Add Guardrails tab to endpoint editor (#22360, @TomeHirata)
• [UI] Add click-to-expand modal for trace attachment images (#22461, @kriscon-db)
• [UI] Enhance Gateway quick start cards with logos, multi-model options, and compact variant (#22513, @xq-yin)
• [Tracing] Add @mlflow/vercel for better Vercel AI SDK tracing in Databricks UC (#22105, @dbrx-euirim)
• [UI] [UI] Refactor API keys page: bulk delete, inline drawer editing, consistent list pattern (#22485, @xq-yin)
• [] Add Guardrail base class and JudgeGuardrail implementation (#21964, @TomeHirata)
• [Tracing] Add multimodal demo traces with synthetic image and audio data (#22465, @kriscon-db)
• [] [FR] #22217 Add configurable provider allow/deny list to AI Gateway (#22291, @etirelli)
• [UI] [UI] Add bulk duplicate and delete actions to AI Gateway endpoints table (#22404, @xq-yin)
• [UI] Redesign endpoint details model section with unified Model card (#22376, @xq-yin)
• [] Add X-MLflow-Gateway-Duration-Ms and X-MLflow-Gateway-Overhead-Duration-Ms response headers (#22229, @PattaraS)
• [UI] Redesign AI Gateway endpoint details: Overview tab, starter code, and Try in Browser (#22284, @xq-yin)
• [] Add guardrail proto definitions, server handlers, and REST store (#21962, @TomeHirata)
• [Build] Add aiohttp as a core dependency of mlflow (#22189, @TomeHirata)
• [] Add gateway guardrail DB layer: schema, store methods, and tests (#21960, @TomeHirata)
• [UI] Add quick-start template cards to AI Gateway empty state (#22249, @xq-yin)
• [Tracing / UI] Add freeform Notes section to trace assessments pane (#21676, @smoorjani)
• [Tracing / UI] [UI] Improve traces table visual design and UX (#20424, @ridgupta26)
• [UI] Improve Image Grid chart UX for multi-run image comparison (#22118, @Rasaboun)
• [Models] Add base_model_path parameter to save PEFT adapter-only with local base model reference (#22052, @rpathade)
• [Tracking] Add mlflow db move-resources command to move resources between workspaces (#21263, @mprahl)
• [Tracking] feat(tracking): Add composite index on metrics (run_uuid, key, step) (#21727, @ravidarbha)

Bug fixes:

• [Tracing / Tracking] Redact presigned URL credentials in urllib3 retry logs (#22995, @B-Step62)
• [Tracing] Fix assessment logging silently dropped in distributed tracing (#22963, @serena-ruan)
• [] Filter _list_budget_windows results by active workspace when request is workspace-scoped (#22885, @copilot-swe-agent)
• [Models / Scoring] Fix uv custom index URLs omitted from model requirements.txt (#22921, @serena-ruan)
• [Tracing / UI] Set mlflow.message.format on gateway passthrough spans to enable Chat tab (#22916, @TomeHirata)
• [Server-infra] Add RBAC regression tests + gateway validator fix (#22862, @PattaraS)
• [Server-infra] Include role-based grants in list_accessible_workspace_names (#22864, @PattaraS)
• [Tracing] Preserve cache_read tokens in claude_code tracing for cache observability (#22683, @dgokeeffe)
• [Tracing] Fix InferenceTableSpanProcessor init for opentelemetry-sdk 1.41.0 (#22867, @harupy)
• [UI] Fix audio artifact viewer failing in non-default workspaces (#22811, @harupy)
• [Tracking] Fix O(n^2) body buffering in WSGIMiddleware for large uploads (#22729, @harupy)
• [Tracking] Propagate workspace_id as X-Databricks-Org-Id header for SPOG support (#22554, @B-Step62)
• [UI] Fix sessions pagination not auto-loading after virtualizer settles (#22808, @daniellok-db)
• [Tracing] Fix _get_token_usage dropping zero-valued token counts (#22748, @copilot-swe-agent)
• [Tracing] Fix _get_token_usage dropping cache token fields (#22818, @harupy)
• [UI] Remove rendering size guards from artifact image and video views (#22816, @kriscon-db)
• [UI] Disable Create Guardrail in AddGuardrailModal until a Guardrail Model endpoint is selected (#22766, @copilot-swe-agent)
• [UI] Disable Guardrails tab in EditEndpointFormRenderer when endpoint.experiment_id is null, refresh i18n messages, and fix JS type-check follow-ups (#22757, @copilot-swe-agent)
• [] Fix OPENAI_API_KEY requirement when guardrails use gateway:/ judge model (#22769, @TomeHirata)
• [Tracking] Fix ABFSS URI double-encoding in _parse_abfss_uri (#22759, @artjen)
• [UI] Persist immediate gateway usage-tracking updates and preserve dirty edits in useEditEndpointForm (#22734, @copilot-swe-agent)
• [] Fix Vertex AI gateway global location handling for Gemini 3 models (#22696, @harupy)
• [Tracing] Fix ValueError: Circular reference detected crash in dump_span_attribute_value (pydantic_ai autolog) (#22693, @barry3406)
• [Tracing] Add Bedrock Converse models to model catalog (#22515, @ajgassner)
• [Tracing] Fix incorrect cost calculation for Anthropic cached requests (#22516, @ajgassner)
• [UI] Disable Create button when experiment name is empty (#22649, @ssam18)
• [Tracing] Fix AsyncHttpxClientWrapper AttributeError by avoiding deepcopy in TraceJSONEncoder (#22742, @harupy)
• [Evaluation] Fix JSON parsing failure for judge model responses containing newlines (#22736, @harupy)
• [Evaluation] Fix non-ASCII characters being escaped in judge template variables (#22728, @harupy)
• [Tracking] Fix pandas 3.x datetime64 resolution compatibility in cast_df_types_according_to_schema (#22705, @copilot-swe-agent)
• [Tracing] Fix conversation simulator adding wrapper span that breaks custom scorers (#21321, @B-Step62)
• [] Fix Bedrock tool-use chat response parsing (#22677, @lavaFreak)
• [UI] Fix AI Gateway provider dropdown by flattening OpenAI / Azure OpenAI (#22701, @harupy)
• [Evaluation / Tracing] Preserve predict_fn signature in simulation turn tracing (#22610, @rogalski)
• [Tracing / Tracking] fix: Extract cached_tokens in OpenAI streaming responses to correct cost tracking (#22620, @Rishabh-git10)
• [Tracing / UI] Auto-select first endpoint and API key in issue detection modal (#22633, @serena-ruan)
• [Model Registry] Use SQL MAX() in create_model_version instead of loading all rows (#22635, @neolunar7)
• [Tracing] Add modality/tool pricing extraction to dev/update_model_catalog.py, backfill anthropic.json / gemini.json, and skip reasoning modality pricing (#22699, @copilot-swe-agent)
• [Tracing] Fix pydantic-ai run_stream_sync autologging broken by AgentSpec forward reference in _returns_sync_streamed_result (#22666, @copilot-swe-agent)
• [UI] Unify execution duration cell rendering in session header rows (#22631, @daniellok-db)
• [] Implement guardrail execution in gateway API handlers (#22306, @TomeHirata)
• [Tracing / Tracking] Fast-path trace-id search in frontend to avoid spans.content scan (#22433, @harupy)
• [UI] Fix tag cutoff in prompts table (#22507, @nananosirova)
• [Scoring] Fix pandas 3.0 StringDtype mismatch in cast_df_types_according_to_schema (#22537, @copilot-swe-agent)
• [Tracing] Fix exception propagation in @mlflow.trace within @trace_disabled context (#22501, @harupy)
• [UI] [UI] Show header and breadcrumb on gateway usage page when no endpoints exist (#22445, @xq-yin)
• [Model Registry] Fix performance degradation for single-tenant installation (#22405, @abaland)
• [UI] Render attachments inline in Details & Timeline span view (#22451, @kriscon-db)
• [UI] Skip markdown rendering for oversized chat content to prevent browser freezes (#22464, @kriscon-db)
• [UI] Render attachment images embedded in complex JSON structures (#22462, @kriscon-db)
• [UI] Handle input_file content type in Responses API chat rendering (#22466, @kriscon-db)
• [UI] Fall back to OpenAI parser for LangChain traces with OpenAI-format data (#22456, @kriscon-db)
• [UI] Combine multi-part Responses API input into a single chat message (#22459, @kriscon-db)
• [UI] Render text content before audio parts in chat messages (#22460, @kriscon-db)
• [Tracing] Handle Gemini SDK bytes repr format in inline_data extraction (#22453, @kriscon-db)
• [UI] Show all output fields alongside attachment renderings in summary view (#22449, @kriscon-db)
• [UI] Handle single dict contents in Gemini chat input normalization (#22455, @kriscon-db)
• [UI] Fix audio player not loading on first render in attachment view (#22450, @kriscon-db)
• [UI] Allow Gemini chat rendering when role field is omitted (#22454, @kriscon-db)
• [Evaluation] Add server side check for registering custom scorer (#22381, @WeichenXu123)
• [UI] [UI] Cache gateway provider/model queries to eliminate redundant fetches (#22440, @xq-yin)
• [Tracing] Eager-load trace relationships in search_traces (#22431, @harupy)
• [Model Registry] Fix cross-workspace copy_model_version failure caused by get_logged_model call (#22262, @TomeHirata)
• [UI] [UI] Fix endpoint details to display all API keys for multi-model endpoints (#22406, @xq-yin)
• [UI] Fix tag cutoff in experiments table (#21871, @nananosirova)
• [Tracking] Fix rename_experiment in SqlAlchemyStore to enforce 500-char name limit (#22418, @copilot-swe-agent)
• [Tracking] Fix download_artifacts and list_artifacts returning wrong status for missing artifacts (#22310, @Karim-siala)
• [Docs / Tracing] Enable async trace logging by default for OSS MLflow (#22304, @PattaraS)
• [UI] Show Save/Cancel buttons only when endpoint has unsaved changes (#22367, @xq-yin)
• [UI] Remove mp4 from AUDIO_EXTENSIONS to fix video artifact preview (#22350, @copilot-swe-agent)
• [Evaluation] Fix MetaPromptOptimizer failing on prompts with no template variables (#22301, @alkispoly-db)
• [Prompts / UI] Fix misleading placeholder text in PromptsListFilters (#22321, @copilot-swe-agent)
• [Evaluation] Fix make_judge with bool/numeric types not reporting aggregated metrics in evaluate() (#22302, @alkispoly-db)
• [UI] Fix dark mode CSS not consistently applied in compare runs page (#22254, @DaoDaoNoCode)
• [Tracking] Fix 500 error in start_trace when merging traces with existing metrics (#22257, @alkispoly-db)
• [Model Registry / Tracing / Tracking] Fix _all_tables_exist to derive expected_tables from Base.metadata dynamically (#22128, @copilot-swe-agent)
• [UI] Hide 'Show more columns' button when runs are selected (#21867, @nananosirova)
• [Evaluation] Support T | None as top-level feedback_value_type in make_judge (#22201, @copilot-swe-agent)
• [UI] Fix dataset details drawer: local source display, profile parsing, broken link (#22143, @Rasaboun)
• [Tracking] Fix --static-prefix not applied to /api/ REST routes (#22159, @TomeHirata)
• [Tracing / Tracking] Map user.id span attribute to mlflow.trace.user during OTel ingestion (#22150, @ajgassner)

Documentation updates:

• [Docs] Docs: Add AI Gateway CLI Tools pages (Claude Code, Codex, Gemini CLI) (#22918, @TomeHirata)
• [Docs / Tracing / UI] Add documentation for Codex CLI, Gemini CLI, and Qwen Code tracing (#22412, @kriscon-db)
• [Docs] Add Live Demo documentation page for demo.mlflow.org (#22840, @serena-ruan)
• [Docs] Docs: Adds Middleware in the tracing section for integrated observability platforms (#22791, @luharshekhar2000-dot)
• [Docs / UI] Rename guardrail stage terminology to Pre-LLM / Post-LLM across guardrail_utils.py, Gateway API, UI, docs, and guardrailValidation.ts (#22767, @copilot-swe-agent)
• [Docs] Fix EvaluationDataset type references in GenAI datasets docs (#22761, @harupy)
• [Docs] docs: add guardrails documentation for AI Gateway (#22577, @TomeHirata)
• [] Allow different PGPORT in docker compose (#22646, @riklopfer)
• [Docs] Remove docs/docs/genai/governance/ai-gateway/legacy/* and redirect legacy AI Gateway routes to ai-gateway/index (#22579, @copilot-swe-agent)
• [Docs] Remove unused parameter from invocation() function example in manual tracing docs (#22524, @prithvipal)
• [Docs] Document uv run --frozen flag for offline/no-network usage in CLAUDE.md (#22505, @copilot-swe-agent)
• [Docs] Add documentation for multimodal trace attachments (#21783, @kriscon-db)
• [Docs] Add AI Gateway Performance & Benchmarks docs page (#22390, @copilot-swe-agent)
• [Docs] Update changelog for 3.11.0 release (#22394, @WeichenXu123)

Small bug fixes and documentation updates:

#22993, #22978, #21482, #21321, #22559, #21319, #22483, #22346, #22091, @B-Step62; #22919, #22925, #22854, #22821, #22786, #22820, #22615, #22819, #22781, #22703, #22565, #22200, #22193, @serena-ruan; #22869, #22863, #22843, #22839, #22837, #22815, #22765, #22762, #22749, #22708, #22486, #22481, #22471, #22401, #22366, #22278, #22281, @harupy; #22848, #22775, #22764, #22763, #22772, #22527, #22435, #22358, #22357, #22356, #22355, #22261, #22259, #22202, #22234, #22243, @TomeHirata; #22824, @vinh412; #22805, #22804, #22803, #22802, #22801, #22800, #22447, #22632, #22735, #22697, #22691, #22508, #22628, #22567, #22482, #22441, #22273, #22272, #22190, #22434, #22154, #22147, #22144, #22268, #22241, #22187, #22074, @daniellok-db; #22557, #22722, #22807, #22721, #22558, #22589, #22319, #22145, #21789, @PattaraS; #22825, #22783, #22789, #22787, #22784, #22782, #22757, #22771, #22770, #22758, #22756, #22755, #22744, #22737, #22715, #22718, #22714, #22713, #22711, #22710, #22707, #22704, #22700, #22692, #22686, #22690, #22689, #22688, #22684, #22685, #22682, #22670, #22669, #22668, #22667, #22664, #22663, #22662, #22655, #22657, #22659, #22601, #22580, #22578, #22576, #22563, #22550, #22549, #22548, #22547, #22545, #22544, #22541, #22540, #22539, #22538, #22511, #22532, #22529, #22528, #22526, #22523, #22525, #22519, #22518, #22517, #22514, #22512, #22509, #22510, #22506, #22504, #22490, #22488, #22480, #22478, #22477, #22470, #22468, #22443, #22438, #22436, #21203, #22417, #22403, #22413, #22399, #22395, #22391, #22392, #22388, #22379, #22348, #22347, #22343, #22341, #22342, #22340, #22339, #22337, #22330, #22329, #22328, #22316, #22309, #22305, #22250, #22207, #22204, #22198, #22177, #22174, @copilot-swe-agent; #22795, @rollyjoel; #22629, #22730, #22614, #22573, #22334, #22502, #22448, #22196, #22263, #22186, #22185, #22184, #21956, #22183, #22246, #21955, @kriscon-db; #22627, #22238, @xsh310; #22295, #22212, @smoorjani; #22463, #22370, #22365, #22240, #22226, @xq-yin; #22473, #22439, #22437, @WeichenXu123; #22387, #21419, @mprahl; #22382, #22373, @SeldonAi; #21848, @alkispoly-db; #22160, @amotl