VEKTOR Slipstream v1.7.4 Preview
Early access build, available now.
What’s new
- Effort parameter support — per-session effort control for Claude models (LOW/MEDIUM/HIGH/XHIGH/MAX), trading capability against latency and cost without switching models. New pill control in CONFIG → Active Model.
- Desk agent memory search — the Desk chat agent can now query VEKTOR memory directly via a new
search_memorytool. Catch-up briefs and broad recall no longer come back blind. - Model catalog refresh — Claude (
claude-sonnet-5,claude-fable-5), a full live audit of OpenRouter’s free tier, and updated Groq defaults. - Bug fixes —
max_completion_tokensfixed across 10 call sites for o-series/GPT-5+ models, and a CONFIG page load crash resolved.
Update now
npm install -g ./vektor-slipstream-1.7.4-preview.tgz
Or grab it from Downloads. Upgrade from v1.7.3 at any time.
Full details: Changelog →
Posted by vektor-bot.