Vellum Assistant

// Overview

Vellum Assistant is an open-source personal AI assistant from Vellum AI, written in TypeScript under the MIT licence. Its README pitches it as an assistant that evolves with you: it learns how you work, remembers what matters and takes action across your apps. It is positioned as a ready-made alternative to assembling a personal AI yourself on top of agent tools such as OpenClaw, Hermes Agent or Claude Code, and is delivered as a desktop app for macOS and Windows with a `vellum` CLI for terminal users.

Memory is the centre of the design. The assistant keeps eight types — episodic, semantic, procedural, emotional, prospective, behavioral, narrative and shared — each with its own staleness window, retrieved with hybrid dense and sparse search and isolated per user and per channel. It extracts structured items such as identity, preferences, projects and events from conversations with source attribution, and runs embeddings locally by default. Its behaviour lives in a SOUL.md file: during onboarding it observes how you communicate and writes its own personality files, keeps a per-user journal of reflections and uses NOW.md as a scratchpad for current focus.

The assistant is also proactive: every hour it re-reads its notes, looks for anything unfinished or due soon and messages you if something needs attention. Security is deny-by-default — actor identity (guardian, trusted or unknown) is resolved once and enforced everywhere, credentials live in a separate process that never reaches the model, and every tool call runs in a sandbox. It can run on Vellum's managed platform or be self-hosted from the same codebase and data model.

// What it does

• Eight memory types with per-type staleness windows, hybrid dense + sparse retrieval, per-user and per-channel isolation, and local embeddings by default
• Self-written identity: a SOUL.md personality built during onboarding, a per-user reflection journal and a NOW.md scratchpad
• Hourly proactive check-ins that surface unfinished or due work on the right channel without interrupting an active conversation
• One assistant and one memory across macOS, Windows, iOS, Web, Voice, Email, Telegram, Slack and Twilio, with OAuth for Slack, Notion, Google, HubSpot, Linear and more
• Permission-gated computer use (files, commands, browser) and SKILL.md + TOOLS.json skills, all sandboxed and deny-by-default
• Works with Anthropic, OpenAI, Google Gemini, Fireworks, OpenRouter, MiniMax and any OpenAI-compatible endpoint, plus local models through Ollama

// Getting started

The desktop app is the project's primary focus; the README also documents a CLI for advanced users, contributors and terminal-based workflows. The installation guide lists macOS Sequoia (15) or later and Windows 10 or later for the desktop app.

bun install -g vellum
vellum hatch

git clone https://github.com/vellum-ai/vellum-assistant.git
cd vellum-assistant
./setup.sh
source ~/.bashrc
vellum hatch

vellum wake        # start services
vellum sleep       # stop services, keep data
vellum client      # interact through the terminal
vellum ps          # view running assistants

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• You want a personal AI assistant that remembers your preferences and projects without maintaining a hand-rolled memory file yourself
• You want one assistant reachable from your desktop, phone, Slack, Telegram and email that picks up a thought where you left it
• You want an assistant that nudges you about unfinished or due work on its own instead of waiting to be asked
• You want to self-host an assistant with sandboxed tools and separated credentials, or run the same codebase on a managed platform

// How Vellum Assistant compares

Vellum Assistant alongside other open-source assistants & chatbots tools AI/TLDR tracks, ranked by GitHub stars.