v1.7.3 Preview is available now — early access build shipping two major additions.
Faraday Security Proxy
A new opt-in security layer sitting between your agent and VEKTOR memory.
- PII interception — detects and redacts personal data before it reaches memory
- Memory poisoning detection — semantic anomaly scoring flags injected or manipulated memories
- Canary tokens — planted in memory to detect and alert on exfiltration attempts
- Taint propagation — tracks compromised data as it flows across memory graph edges
Runs as transparent middleware. No changes to existing MCP tool calls required.
Configurable LLM Model Selection
Model selection is now configurable per-provider across all LLM backends — Claude, OpenAI, Groq, Gemini, and Ollama. Override the default model at runtime or set a persistent preference per provider in config. Affects chat, synthesis, JOT collab, and briefing generation.
How to install
Download from vektormemory.com/downloads and run:
npm install -g ./vektor-slipstream-1.7.3-preview.tgz
Upgrade from v1.7.2 at any time. Full changelog at vektormemory.com/docs/changelog#v173.
Feedback welcome here or in Feature Requests. This is a preview build — report any issues and we’ll iterate fast.