github vellum-ai/vellum-assistant v0.6.5
Vellum 0.6.5

latest releases: v0.8.8, v0.8.8-staging.3, v0.8.8-staging.2...
one month ago

Highlights

  • Overhauled Slack integration with a fully chronological context model: the assistant now renders channel and DM history with thread tags, timestamps, edited/deleted markers, and inline reactions, and the OAuth setup flow has been simplified to a one-click install with an automated bot+user token capture.
  • Personal Knowledge Base (PKB) is now fully integrated into the assistant loop, with automatic indexing on file writes and memory saves, hybrid dense/sparse search, relevance-filtered injection hints surfaced in the system reminder, and a new memory compact CLI command for backfilling long nodes.
  • Major security hardening against prompt injection: web search results, messaging tool output, and other external content are now wrapped in explicit trust boundaries, and a new data-driven bash risk classifier (with approval policy layer) is being rolled out to gate risky shell commands.
  • Google Meet participation is now available, allowing the assistant to join meetings as a bot via a Chrome extension bridge, send and read chat, transcribe audio with STT, speak responses with TTS lip-sync, and optionally show a talking-head avatar on a virtual camera.
  • New xAI speech and voice support, LaTeX math rendering on macOS, and broader model updates: xAI has been added as both a TTS and real-time STT provider, block-level LaTeX math is now rendered natively in the macOS chat view, and the default model has been updated to Claude Opus 4.7 with expanded OpenRouter and Gemini 2.5 catalog support.

Build: 0.6.5
Commit: 2e09d9721
Built at: 2026-04-22 00:16:39 UTC

Don't miss a new vellum-assistant release

NewReleases is sending notifications on new releases.