Skip to main content

v1.81.6 - Logs v2 with Tool Call Tracing

Krrish Dholakia
Krrish Dholakia
CEO, LiteLLM
Ishaan Jaff
Ishaan Jaff
CTO, LiteLLM

Deploy this version​

docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:main-v1.81.6

Key Highlights​

Logs View v2 with Tool Call Tracing - Redesigned logs interface with side panel, structured tool visualization, and error message search for faster debugging.

Let's dive in.

Logs View v2 with Tool Call Tracing​

This release introduces comprehensive tool call tracing through LiteLLM's redesigned Logs View v2, enabling developers to debug and monitor AI agent workflows in production environments seamlessly.

This means you can now onboard use cases like tracing complex multi-step agent interactions, debugging tool execution failures, and monitoring MCP server calls while maintaining full visibility into request/response payloads with syntax highlighting.

Developers can access the new Logs View through LiteLLM's UI to inspect tool calls in structured format, search logs by error messages or request patterns, and correlate agent activities across sessions with collapsible side panel views.

Get Started

New Models / Updated Models​

New Model Support​

ProviderModelContext WindowInput ($/1M tokens)Output ($/1M tokens)Features
AWS Bedrockamazon.nova-2-pro-preview-20251202-v1:01M$2.19$17.50Chat completions, vision, video, PDF, function calling, prompt caching, reasoning
Google Vertex AIgemini-robotics-er-1.5-preview1M$0.30$2.50Chat completions, multimodal (text, image, video, audio), function calling, reasoning
OpenRouteropenrouter/xiaomi/mimo-v2-flash262K$0.09$0.29Chat completions, function calling, reasoning
OpenRouteropenrouter/moonshotai/kimi-k2.5---Chat completions
OpenRouteropenrouter/z-ai/glm-4.7202K$0.40$1.50Chat completions, vision, function calling, reasoning

Features​

  • AWS Bedrock

    • Messages API Bedrock Converse caching and PDF support - PR #19785
    • Translate advanced-tool-use to Bedrock-specific headers for Claude Opus 4.5 - PR #19841
    • Support tool search header translation for Sonnet 4.5 - PR #19871
    • Filter unsupported beta headers for AWS Bedrock Invoke API - PR #19877
    • Nova grounding improvements - PR #19598, PR #20159

  • Anthropic

    • Remove explicit cache_control null in tool_result content - PR #19919
    • Fix tool handling - PR #19805

  • Google Gemini / Vertex AI

    • Add Gemini Robotics-ER 1.5 preview support - PR #19845
    • Support file retrieval in GoogleAIStudioFilesHandle - PR #20018
    • Add /delete endpoint support - PR #20055
    • Add custom_llm_provider as gemini translation - PR #19988
    • Subtract implicit cached tokens from text_tokens for correct cost calculation - PR #19775
    • Remove unsupported prompt-caching-scope-2026-01-05 header for vertex ai - PR #20058
    • Add disable flag for anthropic gemini cache translation - PR #20052
    • Convert image URLs to base64 in tool messages for Anthropic on Vertex AI - PR #19896

  • xAI

    • Add grok reasoning content support - PR #19850
    • Add websearch params support for Responses API - PR #19915
    • Add routing of xai chat completions to responses when web search options is present - PR #20051
    • Correct cached token cost calculation - PR #19772

  • Azure OpenAI

    • Use generic cost calculator for audio token pricing - PR #19771
    • Allow tool_choice for Azure GPT-5 chat models - PR #19813
    • Set gpt-5.2-codex mode to responses for Azure and OpenRouter - PR #19770

  • OpenAI

    • Fix max_input_tokens for gpt-5.2-codex - PR #20009
    • Fix gpt-image-1.5 cost calculation not including output image tokens - PR #19515

  • Hosted VLLM

    • Support thinking parameter in anthropic_messages() and .completion() - PR #19787
    • Route through base_llm_http_handler to support ssl_verify - PR #19893
    • Fix vllm embedding format - PR #20056

  • OCI GenAI

    • Serialize imageUrl as object for OCI GenAI API - PR #19661

  • Volcengine

    • Add context for volcengine models (deepseek-v3-2, glm-4-7, kimi-k2-thinking) - PR #19335

  • Chinese Providers

    • Add prompt caching and reasoning support for MiniMax, GLM, Xiaomi - PR #19924

  • Vercel AI Gateway

Bug Fixes​

  • Google

    • Fix gemini-robotics-er-1.5-preview entry - PR #19974

  • General

    • Fix output_tokens_details.reasoning_tokens None - PR #19914
    • Fix stream_chunk_builder to preserve images from streaming chunks - PR #19654
    • Fix aspectRatio mapping in image edit - PR #20053
    • Handle unknown models in Azure AI cost calculator - PR #20150

  • GigaChat

    • Ensure function content is valid JSON - PR #19232

LLM API Endpoints​

Features​

Bugs​

  • General
    • Fix token calculations and refactor - PR #19696

Management Endpoints / UI​

Features​

  • Proxy CLI Auth

    • Add configurable CLI JWT expiration via environment variable - PR #19780
    • Fix team cli auth flow - PR #19666

  • Virtual Keys

    • UI: Auto Truncation of Table Values - PR #19718
    • Fix Create Key: Expire Key Input Duration - PR #19807
    • Bulk Update Keys Endpoint - PR #19886

  • Logs View

  • Models + Endpoints

    • Add sortBy and sortOrder params for /v2/model/info - PR #19903
    • Fix Sorting for /v2/model/info - PR #19971
    • UI: Model Page Server Sort - PR #19908

  • Usage & Analytics

    • UI: Usage Export: Breakdown by Teams and Keys - PR #19953
    • UI: Usage: Model Breakdown Per Key - PR #20039

  • UI Improvements

    • UI: Allow Admins to control what pages are visible on LeftNav - PR #19907
    • UI: Add Light/Dark Mode Switch for Development - PR #19804
    • UI: Dark Mode: Delete Resource Modal - PR #20098
    • UI: Tables: Reusable Table Sort Component - PR #19970
    • UI: New Badge Dot Render - PR #20024
    • UI: Feedback Prompts: Option To Hide Prompts - PR #19831
    • UI: Navbar: Fixed Default Logo + Bound Logo Box - PR #20092
    • UI: Navbar: User Dropdown - PR #20095
    • Change default key type from 'Default' to 'LLM API' - PR #19516

  • Team & User Management

    • Fix /team/member_add User Email and ID Verifications - PR #19814
    • Fix SSO Email Case Sensitivity - PR #19799
    • UI: Internal User: Bulk Add - PR #19721

  • AI Gateway Features

    • Add support for making silent LLM calls without logging - PR #19544
    • UI: Fix MCP tools instructions to display comma-separated strings - PR #20101

Bugs​

  • Fix Model Name During Fallback - PR #20177
  • Fix Health Endpoints when Callback Objects Defined - PR #20182
  • Fix Unable to reset user max budget to unlimited - PR #19796
  • Fix Password comparison with non-ASCII characters - PR #19568
  • Correct error message for DISABLE_ADMIN_ENDPOINTS - PR #19861
  • Prevent clearing content filter patterns when editing guardrail - PR #19671
  • Fix Prompt Studio history to load tools and system messages - PR #19920
  • Add WATSONX_ZENAPIKEY to WatsonX credentials - PR #20086
  • UI: Vector Store: Allow Config Defined Models to Be Selected - PR #20031

Logging / Guardrail / Prompt Management Integrations​

Features​

  • DataDog

    • Add agent support for LLM Observability - PR #19574
    • Add datadog cost management support and fix startup callback issue - PR #19584
    • Add datadog_llm_observability to /health/services allowed list - PR #19952
    • Check for agent mode before requiring DD_API_KEY/DD_SITE - PR #20156

  • OpenTelemetry

    • Propagate JWT auth metadata to OTEL spans - PR #19627
    • Fix thread leak in dynamic header path - PR #19946

  • Prometheus

    • Add callbacks and labels - PR #19708
    • Add clientip and user agent in metrics - PR #19717
    • Add tpm-rpm limit metrics - PR #19725
    • Add model_id label to metrics - PR #19678
    • Safely handle None metadata in logging - PR #19691
    • Resolve high CPU when router_settings in DB by avoiding REGISTRY.collect() - PR #20087

  • Langfuse

    • Add litellm_callback_logging_failures_metric for Langfuse, Langfuse Otel and other Otel providers - PR #19636

  • General Logging

    • Use return value from CustomLogger.async_post_call_success_hook - PR #19670
    • Add async_post_call_response_headers_hook to CustomLogger - PR #20083
    • Add mock client factory pattern and mock support for PostHog, Helicone, and Braintrust integrations - PR #19707

Guardrails​

  • Presidio

    • Reuse HTTP connections to prevent performance degradation - PR #19964

  • Onyx

  • General

    • Add guardrail model argument feature - PR #19619
    • Fix guardrails issues with streaming-response regex - PR #19901
    • Remove enterprise requirement for guardrail monitoring (docs) - PR #19833

Spend Tracking, Budgets and Rate Limiting​

  • Add event-driven coordination for global spend query to prevent cache stampede - PR #20030

Performance / Loadbalancing / Reliability improvements​

  • Resolve high CPU when router_settings in DB - by avoiding REGISTRY.collect() in PrometheusServicesLogger - PR #20087
  • Reuse HTTP connections in Presidio - to prevent performance degradation - PR #19964
  • Event-driven coordination for global spend query - prevent cache stampede - PR #20030
  • Fix recursive Pydantic validation issue - PR #19531
  • Refactor argument handling into helper function to reduce code bloat - PR #19720
  • Optimize logo fetching and resolve MCP import blockers - PR #19719
  • Improve logo download performance using async HTTP client - PR #20155
  • Fix server root path configuration - PR #19790
  • Refactor: Extract transport context creation into separate method - PR #19794
  • Add native_background_mode configuration to override polling_via_cache for specific models - PR #19899
  • Initialize tiktoken environment at import time to enable offline usage - PR #19882
  • Improve tiktoken performance using local cache in lazy loading - PR #19774
  • Fix timeout errors in chat completion calls to be correctly reported in failure callbacks - PR #19842
  • Fix environment variable type handling for NUM_RETRIES - PR #19507
  • Use safe_deep_copy in silent experiment kwargs to prevent mutation - PR #20170
  • Improve error handling by inspecting BadRequestError after all other policy types - PR #19878

Database Changes​

Schema Updates​

TableChange TypeDescriptionPRMigration
LiteLLM_ManagedVectorStoresTableNew ColumnsAdded team_id and user_id fields for permission managementPR #19972Migration

Migration Improvements​

  • Fix Docker: Use correct schema path for Prisma generation - PR #19631
  • Resolve 'relation does not exist' migration errors in setup_database - PR #19281
  • Fix migration issue and improve Docker image stability - PR #19843
  • Run Prisma generate as nobody user in non-root Docker container for security - PR #20000
  • Bump litellm-proxy-extras version to 0.4.28 - PR #20166

Documentation Updates​

Infrastructure / Testing Improvements​

  • Add test coverage for Router.get_valid_args and improve code coverage reporting - PR #19797
  • Add validation of model cost map as CI job - PR #19993
  • Add Realtime API benchmarks - PR #20074
  • Add Init Containers support in community helm chart - PR #19816
  • Add libsndfile to main Dockerfile for ARM64 audio processing support - PR #19776

New Contributors​

Full Changelog: https://github.com/BerriAI/litellm/compare/v1.81.3.rc...v1.81.6