Vellum Assistant
A personal AI assistant with eight kinds of memory that learns how a user works, remembers what matters, and takes action across their apps and channels, available managed or self-hosted.
Vellum Assistant is a personal AI assistant meant to reach usable results faster than assembling one from scratch on OpenClaw, Hermes Agent, or Claude Code. It has eight memory types (episodic, semantic, procedural, emotional, prospective, behavioral, narrative, and shared), each with its own staleness window and hybrid dense and sparse retrieval, and extracts structured items such as identity, preferences, projects, and events from conversations with source attribution and deduplication; embeddings run locally by default.
Its behavior lives in a SOUL.md file that the assistant writes for itself during onboarding by observing how the user communicates, alongside a per-user journal and a NOW.md scratchpad for current focus. Every hour it re-reads its notes and proactively messages the user about anything unfinished, without interrupting an active conversation.
It resolves actor identity before granting memory or tool access, runs every tool call in a sandbox, and supports macOS, Windows, iOS, web, voice, email, Telegram, Slack, and Twilio channels, hosted on the Vellum Platform or self-hosted.