Vellum Assistant is a full assistant system that remembers preferences, tracks ongoing work, and can take action through connected apps and channels. The repo covers the assistant runtime, gateway, CLI, and client apps, plus the files that shape its behavior and permissions.
Builders who want an assistant that remembers them, works across channels, and can take actions for them.
You can use one assistant across devices and channels without re-explaining your context every time.
Stores episodic, semantic, procedural, emotional, prospective, behavioral, narrative, and shared memory with source attribution and deduplication.
Re-reads notes on a schedule, checks for unfinished or due items, and sends the right notification without interrupting active chats.
Works across macOS, iOS, web, voice, email, Telegram, Slack, and Twilio with one shared assistant and one shared memory.
Runs tool calls in a sandbox, gates permissions, and keeps credentials in a separate process.
Loads skills from `SKILL.md` and `TOOLS.json`, then adds tools and prompt sections at runtime.
Connects to Anthropic, OpenAI, Gemini, Fireworks, OpenRouter, MiniMax, Atlas Cloud, and local Ollama models.
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 vellum terminal # open a shell into a managed assistant container vellum upgrade # upgrade to latest version
A personal AI assistant that evolves with you.
8 different types of memory (episodic, semantic, procedural, emotional, prospective, behavioral, narrative, shared) make it truly yours. It learns how you work, remembers what matters, and takes action across your apps.
If you've set up a Personal AI on OpenClaw, Hermes Agent, or Claude Code, you know how long it takes, and how many times you have to hatch a new one to get it right. Vellum gets you the result you're looking for out of the box, one download away.
| Area | Summary |
|---|---|
| Memory | Eight types (episodic, semantic, procedural, emotional, prospective, behavioral, narrative, shared), each with its own staleness window, hybrid dense + sparse retrieval, and per-user and per-channel isolation. Structured items (identity, preferences, projects, events) extracted from conversations with source attribution and dedup. Embeddings run locally by default. Not a SQLite + Markdown file you maintain yourself. |
| Identity | Behavior lives in SOUL.md. During onboarding the assistant observes how you communicate and writes its own personality files. It keeps a per-user journal of reflections and uses NOW.md as a scratchpad for current focus and active threads. |
| Proactivity | Every hour the assistant re-reads its notes, looks for anything unfinished or due soon, and messages you if something needs attention. Notifications go to the right channel and won't interrupt an active conversation. |
| Security | Actor identity (guardian, trusted, unknown) is resolved once and enforced everywhere; unknown actors can't read memory, trigger tools, or escalate. Credentials live in a separate process and never reach the model. Every tool call runs in a sandbox. The default is to deny. |
| Channels | macOS, iOS, Web, Voice, Email, Telegram, Slack, Twilio. One assistant, one memory, every channel. |
| OAuth | Slack, Notion, Google, HubSpot, Linear, Discord, Twitter, Telegram, Twilio. No hand-rolled token refresh. |
| Hosting | Managed runtime on Vellum Platform, or self-hosted. Same codebase, same data model. |
1. Sign up or download the app
2. Pick your mode
3. Hatch your assistant
Prefer the terminal? See CLI install below.
The CLI works but the desktop app is our primary focus. Available for advanced users, contributors, and non-macOS environments.
Install
bun install -g vellum
vellum hatch
Install from source
git clone https://github.com/vellum-ai/vellum-assistant.git
cd vellum-assistant
./setup.sh
source ~/.bashrc
vellum hatch
Common commands
vellum wake # start services
vellum sleep # stop services, keep data
vellum client # interact through the terminal
vellum ps # view running assistants
vellum terminal # open a shell into a managed assistant container
vellum upgrade # upgrade to latest version
All commands target the default assistant. If you have multiple, pass the assistant ID as the second argument.
Guides
| Area | Summary |
|---|---|
| Computer use | The assistant works in its own sandbox, and with your approval reaches your actual machine: reads and edits files, runs commands, drives the browser. Every action is permission-gated, and you can grant once, for ten minutes, or always. |
| Skills | Plugins defined by a SKILL.md and a TOOLS.json that add tools and prompt sections at runtime, sandboxed like everything else. Install them from the catalog, bundle them, or drop them in the workspace. |
| Channels | One assistant with one memory, reachable from the macOS app, Telegram, or Slack. Start a thought in one channel and pick it up in another. |
| Multi-provider support | Works with Anthropic, OpenAI, Google Gemini, Fireworks, OpenRouter, MiniMax, Atlas Cloud, and any OpenAI-compatible endpoint. Local models run through Ollama. Embeddings run on local ONNX by default and fall back to cloud providers automatically. |
The canonical sources for who we are and how we talk about what we're building. The docs site at vellum.ai/docs is a rendered view of these files.
| Doc | What it is |
|---|---|
| Constitution | Who we are, what we believe, and what we refuse to compromise on |
| Glossary | The shared vocabulary we use to talk about personal intelligence |
| Section | What's covered |
|---|---|
| Architecture | Platform domains, repo structure, runtime · clients · gateway |
| Security & Permissions | Sandbox, credentials, trust rules, permission modes |
| Features & Capabilities | Integrations, dynamic skills, browser, attachments, media embeds |
| API & Communication | SSE event stream, event payloads, remote access |
| Development Workflow | Claude Code commands, parallel PRs, review loops, release pipeline |
We welcome contributions from everyone.
MIT. See License. Integration logos from Simple Icons, licensed CC0 1.0.
Vellum Assistant is open-source software built by Vellum AI, a for-profit company. We also offer a managed product, the Vellum Platform, which sustains the business. Free to use and modify under MIT, and we're committed to keeping it that way.
Built with 💚 by Vellum
Sign in to join the discussion.
No comments yet. Be the first to say what this is good for.
🎨 Best DeepSeek Harness Design Plugin. The open-source Claude Design alternative. 🖥️ Local-first desktop app. 🖼️ Your coding agent becomes the design engine: prototypes, landing pages, dashboards, slides, images & video — real files, HTML/PDF/PPTX/MP4 export. 🤖 Claude Code / Codex / Cursor / DeepSeek Harness / OpenCode & 20+ CLIs via BYOK.
"Vibe-Trading: Your Personal Trading Agent"
OpenHuman is an open source personal AI for Mac, Windows and Linux — local-first memory, agent orchestration, and deep research.
Your Personal AI Assistant; easy to install, deploy on your own machine or on the cloud; supports multiple chat apps with easily extensible capabilities.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
macOS video editor built for AI