LiteLLM v1.106.0-rc.1
LiteLLM v1.106.0-rc.1 fixes a security issue where LiteLLM caller keys were forwarded to MCP servers, adds decision model integrations, and backports several UI and guardrail features.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
LiteLLM released v1.106.0-rc.1 with key security fixes, features, and UI updates. Notable changes include: - fix(mcp): Ensures LiteLLM caller keys are never forwarded to Model Context Protocol (MCP) servers, and adds request-local guardrail bearers. - feat(decisions): Backports Microsoft-Decision-1, decision models Admin UI, decision model guardrails, and decision auto-router setup. - feat(guardrails): Splits logging-only scope. - UI & Admin: Updates including the System One request form, redesigned Decisions playground, Home page dashboard root, and credential edit fixes. - Supply Chain Security: Provides cosign verification instructions using pinned commit hashes or protected release tags.
可执行摘要
LiteLLM release candidate v1.106.0-rc.1 patches an MCP credential leak preventing caller key forwarding, introduces Microsoft-Decision-1 and decision model support, and updates guardrail handling.
- Agent 实用度
- 77/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade to v1.106.0-rc.1 if utilizing MCP integrations to prevent unintended leakage of caller LiteLLM keys to MCP servers.
- Verify container image integrity via cosign using the pinned commit hash or release tag before deploying the RC build.