Changelog

What's new in Cognisafe

New features, security improvements, and platform updates — shipped continuously. Follow along or read the docs for implementation details.

v0.9.0

Feature releaseMay 2026

newReference architectures — 8 production patterns: HITL approval, MCP tool governance, air-gapped, APIM insertion, SIEM integration, SOC integration, AKS production, Semantic Kernel
newLM Studio provider support — monitor local models during development with zero config
newvLLM, NVIDIA NIM, and Hugging Face TGI provider docs and proxy support
newMulti-agent architecture diagram on the homepage
improvedDeployment blueprint docs: AKS, Kubernetes Helm, APIM, Azure Sentinel, Grafana/Prometheus (8 pages)

Feature releaseApril 2026

newSOC 2 Type II evidence pack — live API endpoint and one-click PDF export for compliance audits
newCustom scorer definitions — write LLM-as-judge prompts, regex, or keyword list scorers and attach them to any project
newPublic Python SDK on PyPI (pip install cognisafe)
newPublic Node.js SDK on npm (npm install cognisafe)
newGitHub Actions integration — block PRs when red-team runs detect regressions
improvedGovernance dashboard: added SOC 2 tab alongside Overview and Red Team
improvedSafety worker now runs custom scorers (keyword_list, regex, prompt_based) asynchronously alongside built-in PyRIT scorers

Security releaseMarch 2026

newDatabase security controls: TLS enforcement, email encryption, 90-day data retention purge
newHTTP security headers added to all Next.js responses (CSP, HSTS, X-Frame-Options)
newpgaudit tamper-evident audit trail for enterprise deployments
improvedHMAC-SHA256 webhook signing on all outbound payloads

Feature releaseFebruary 2026

newLikert severity scale (1–5) for all safety scores — replaces pass/fail binary
new3 new OWASP LLM scorers: Supply Chain, Data Poisoning, Vector & Embedding Weaknesses
newAlerting: Slack, email, and webhook notifications on safety score thresholds
newAgent tagging — assign named API keys per agent for per-agent attribution in the dashboard
newIntegration pages for CrewAI, LangGraph, AutoGen, Semantic Kernel, OpenClaw, ZeroClaw

Feature releaseJanuary 2026

newAutomated red team campaigns using PyRIT with TAP (Tree of Attacks with Pruning)
newGovernance dashboard — RBAC, member management, audit log, compliance evidence
newOutbound webhooks for safety events — filter by scorer and severity
newSelf-hosting docs: Docker Compose, Kubernetes Helm, environment variables reference

Feature releaseDecember 2025

newFull OWASP LLM Top 10 coverage — all 10 categories with async scoring
newMCP tool call monitoring — intercept and score Model Context Protocol server invocations
newInter-agent message inspection for multi-agent pipelines
newCompliance mockup in dashboard: framework coverage, evidence export

Beta launchNovember 2025

Want to see what's coming next?