Skip to content

v1.1 — The Full Arsenal+

Choose a tag to compare

View all tags
@ankitlade12 ankitlade12 released this 24 Mar 14:11
· 255 commits to main since this release
v1.1
55c4576

What's New

A major feature release building on v1.0.0 with the final safety modules.

New Modules

  • Toxicity & Content Safety Filter — Block harmful content across 7 categories with pattern-based and ML modes (#20)
  • Chain-of-Thought Auditor — Audit agent reasoning traces for misalignment (#19)
  • Hallucination / Grounding Guard — Catch hallucinations via grounding checks (#18)
  • Code Safety Shield — Scan LLM-generated code for insecure patterns (#17)
  • MCP Server Security — Validate MCP server trust and tool call policies (#16)
  • Multi-Agent Graph Safety — Budget propagation through agent hierarchies (#15)
  • Google Gemini Support — Zero-code-change patching for Gemini SDK (#14)
  • ML-Powered Injection Shield — TF-IDF classifier for sophisticated jailbreaks (#13)
  • Model Downgrade Cascade — Auto-switch to cheaper models as budget depletes (#9)
  • Semantic Dedup (Replay Shield) — Duplicate prompt detection (#10)
  • Cost Attribution Tags — Per-tag cost breakdowns (#11)
  • Tool-Call Firewall — Allow/block list on LLM tool calls (#7)
  • Canary Token Injection — Detect system prompt leakage (#8)
  • Provider-Aware Cost Analytics — Per-provider spend tracking (#6)
  • Latency Circuit Breaker — Kill slow calls before they cascade (#12)

Full Changelog: v0.3.0...v1.1

Assets 2
Loading