LiteLLM v1.109.0-dev.2
LiteLLM v1.109.0-dev.2 introduces breaking changes across proxy customer pagination, router fallback behaviors, and legacy fallback shims, alongside new features like the 'ai&' provider and per-model RPM/TPM project quota enforcement.
Why this signal matters
LiteLLM v1.109.0-dev.2 is a pre-release containing multiple breaking modifications, core fixes, and ecosystem updates. Notably, breaking changes affect proxy API pagination (customer lists and usage filters), router fallback mechanics (explicit allowed fails behavior and child tier pin isolation), and the removal of legacy extends/mixed-rule fallback shims. Features include the addition of the 'ai&' (aiand) provider, per-provider and per-model RPM/TPM quota enforcement in the proxy, and improved support for MCP (Model Context Protocol) tool rounds in streaming follow-up requests. The release also includes Rust core refactoring (reqwest upgrade to 0.13, moving MCP out of litellm-http), image signature verification via Cosign, and provider-specific adjustments for Bedrock, Anthropic, Cohere, Vertex AI, HuggingFace TEI, and Ollama.
Actionable summary
LiteLLM v1.109.0-dev.2 includes breaking changes: pagination for proxy customer lists/usage filters, modified router cooldown and failure counting rules, child tier pin isolation, and removal of legacy fallback shims. It also adds the ai& provider, per-provider/model RPM/TPM enforcement, and updates MCP streaming response handling.
- Agent usefulness
- 94/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Update proxy integrations to handle paginated customer lists and usage filter endpoints.
- Review router configurations for compatibility with isolated child tier pins and updated fallback deployment cooldown logic.
- Remove any remaining usage of the legacy extends and mixed-rule shims in fallback configs.
- Verify Docker image signatures using the published commit hash 'e82a3c6' or release tag via Cosign if deploying container images directly.