v0.30.0 v1.7.20
vllm-project/vllmSep 22, 2026 → Sep 22, 20260 days between releases
What's New
Adds TypeSafe Jev integration for human-in-the-loop tool call auditing.
Breaking Changes
- Tool calls no longer rely on chat model JSON output for structured approval.
New Features
- TypeSafe Jev integration as a HITL audit backend for agent tool calls.
- Structured pass/block scoring for non-allowlisted tool calls.
- Built-in rules for authorized pentesting and blocking destructive operations.
- Custom organization policy evaluated as an overlay on built-in rules.
Improvements
- Parallel risk assessment for tool calls.
- Local rule synthesis reduces reliance on chat models.
Upgrade Notes
Configure the Jev backend for non-allowlisted tool calls. Define custom organization policies as overlays. The OpenAI-compatible reviewer remains available.
The Two Releases
v0.30.0
Sep 22, 2026· 30 days ago
This release introduces Model Runner V2 with significant performance optimizations and CUDA graph improvements, adds support for new models like DeepSeek-V4.1-Flash and GLM-5.3-Flash, and enhances large-scale serving capabilities with PCP+DCP and advanced KV offloading. It also introduces new quantization strategies and watermarking features.
Full release notesSep 22, 2026· 30 days ago
Integrates TypeSafe Jev as a human-in-the-loop audit backend. Tool calls are sent to Jev for structured approval/rejection based on risk assessment.
Full release notes