Model BehaviorAutomated release watch

LiteLLM v1.104.0-rc.1

LiteLLM v1.104.0-rc.1 is a release candidate featuring a breaking migration of the Langfuse SDK callback to v4, Docker image cosign verification, and multiple fixes across routing, Presidio streaming, and proxy callbacks.

Human read

Why this signal matters

LiteLLM released v1.104.0-rc.1 with signed Docker images via cosign and instructions for verification. The release includes a breaking change migrating the SDK callback to Langfuse v4 (PR 36741). Other notable changes include awaiting the Redis budget pipeline before sync reads, counting team unified access group MCP servers in key grant validation, fixing Presidio output masking during initial keepalives/split characters in streamed /v1/messages, unregistering removed logging callbacks in the proxy, reverting the top-N key cap and daily global spend rollup, and adding OpenAI cached image input prices to the cost map.

Agent parse

Actionable summary

LiteLLM v1.104.0-rc.1 introduces a breaking change by migrating Langfuse callbacks to SDK v4. It also includes key bug fixes for Presidio streaming masking, Redis budget pipelines, MCP server key validation, and proxy logging callback unregistration.

Agent usefulness
89/100
Confidence
96%
Canonical data
JSON + Markdown
Next actions

What builders should check

  • Upgrade to Langfuse v4 if relying on the LiteLLM Langfuse SDK callback integration.
  • Verify Docker image signatures using cosign against the pinned commit hash or release tag if pulling container images directly.
  • Audit proxy and budget configurations if relying on daily global spend rollup or top-N key caps, which were reverted in this release candidate.
Classification

Tags and routing

gatewaymodelsobservability
Related signals

Continue the thread