Vellum Docs

Vellum is a personal AI assistant that remembers, learns, and takes real action across your apps, files, and communication channels.

Vellum runs in Vellum Cloud by default, can run locally or on your own infrastructure, and ships as open-source software. These docs cover setup, memory, channels, skills, tools, hosting, security, and developer workflows.

Install: curl -fsSL https://www.vellum.ai/install.sh | bash && . ~/.config/vellum/env

Repo: https://github.com/vellum-ai/vellum-assistant

Docs: https://www.vellum.ai/docs

Full LLM index: https://www.vellum.ai/llms.txt

Every page below is served as Markdown at its canonical URL with a .md suffix (e.g. https://www.vellum.ai/docs/getting-started.md) or by sending an Accept: text/markdown header.

Docs

• Meet Vellum: Vellum documentation — guides, tutorials, and references for building your personal AI assistant.
• Vellum Constitution: The Vellum Constitution — our purpose, worldview, principles, and the commitments that guide every product, design, and engineering decision we make.
• Developer Guide: Technical reference for contributors and developers working on the Vellum Assistant platform.
• API & Communication: SSE event stream, event payloads, connection management, and remote access via SSH.
• Architecture: Platform domains, repository structure, and how the runtime, clients, and gateway fit together.
• Contributing: How to set up the Vellum Assistant repo locally, run tests, follow code conventions, and submit a pull request.
• Development Workflow: Claude Code slash commands, parallel PR execution, automated review loops, and the release pipeline.
• Features & Capabilities: Integrations, dynamic skill authoring, browser automation, attachments, and media embeds.
• Get Started: Developer on-ramp for Vellum Assistant: repo layout, local dev setup, CLI, HTTP API, and SSE event stream.
• Security & Permissions: Sandbox model, credential storage, trust rules, and permission modes for the Vellum Assistant platform.
• Environments: Choose a deployment environment for Vellum: Vellum Cloud (recommended), local, or user-hosted (GCP or a custom Linux host), with trade-offs for each.
• Extensibility: All of the ways your Assistant can build on themselves. Overview of the surfaces a plugin can bundle and how they compose.
• Apps: Apps let a plugin ship persistent, interactive UI (dashboards, tools, games) served in the workspace panel, built as compiled React.
• Channels: Incoming webhooks can be accepted by declaring public ingress in channels/ingress.json. The gateway signature-checks those routes and forwards them to matching plugin handlers.
• Distribution: How plugins ship: a curated marketplace catalog, installing by name from the CLI, installing directly from a GitHub URL (untrusted), the marketplace.json manifest, and why entries pin an immutable commit.
• Hooks: Lifecycle hooks let a plugin run code at fixed points during the Assistant's lifecycle.
• MCP servers: A plugin declares MCP servers in a root mcp.json. The assistant connects them on install and registers their tools alongside workspace-configured MCP tools.
• Plugins: How a plugin is laid out on disk, what its package.json manifest declares, and the single @vellumai/plugin-api package every surface imports from.
• Routes: Routes let a plugin serve HTTP endpoints in its own /x/plugins// namespace. Public internet webhooks are declared separately as plugin ingress.
• Skills: Skills let a plugin bundle instructions, assets, and scripts the Assistant loads on demand when a conversation matches what the skill is for.
• Tools: Tools let a plugin add new actions the model can call, landing in the same catalog as the Assistant's built-in tools.
• Getting Started: Get started with Vellum: installation, key concepts, a quick start guide, and self-improving skills.
• Installation: Get started with Vellum: sign up for Vellum Cloud, install the iOS or Android app, the macOS or Windows desktop app, or self-host. System requirements, setup, and permissions.
• Quick Start: Your first 2 minutes with Vellum — setup, meeting your assistant, and getting started.
• What is Vellum?: What is Vellum? A personal AI assistant with tools, memory, identity, and a private cloud workspace, different from ChatGPT or Claude.
• Your First Skill: Use and create your first Vellum skill — discover built-in skills, install new ones, and build custom skills.
• Glossary: Vellum glossary — shared definitions for assistant, guardian, personal intelligence, memory, skills, channels, gateway, and other key terms.
• Help: Get help with Vellum — frequently asked questions, common issues, and support channels.
• Common Issues: Troubleshoot common Vellum issues — installation, permissions, skills, performance, and workspace problems.
• FAQ: Frequently asked questions about Vellum — product capabilities, privacy, data handling, and your assistant.
• Getting Help: How to get help with Vellum — community, GitHub issues, email support, and what information to include.
• Hosting options: Choose where your assistant runs: Vellum Cloud (recommended), local on your Mac or Windows PC, or self-hosted on your own infrastructure.
• Cloud hosting: Run your assistant on Vellum Cloud: always-on, sandboxed per account, reachable from web, desktop, voice, and chat channels.
• GCP: Run your assistant on a Compute Engine VM in your own Google Cloud project: provision the VM, hatch on it, and reach it through a tunnel.
• Local hosting: Run your assistant locally on your Mac or Windows PC with setup guidance, tradeoffs, and best practices.
• Pair a device: Reach a self-hosted assistant from your phone, tablet, or another computer: open a tunnel, pair a device, and revoke access when you're done.
• Key Concepts: Core Vellum concepts — workspace, skills and tools, memory and context, channels, assistant identity, and glossary.
• Agentic Payments: How your assistant pays for things on your behalf through your Link account: approve an amount in the Link app, and it checks out with a single-use virtual card.
• Channels: Vellum communication channels — desktop app, CLI, Telegram, Slack, email, and phone. Same assistant everywhere, adapted to each channel.
• Memory & Context: How Memory v3 captures, organizes, connects, and recalls useful knowledge across conversations, plus the controls for creating and refining memories.
• Model Profiles: Control which LLM model and settings your assistant uses for each job — conversations, memory, background tasks — and optimize token costs with profiles and call-site overrides.
• OAuth Integrations: How Vellum connects to third-party services via OAuth2: supported services, how the credential vault works, security model, and troubleshooting.
• Scheduling: How scheduling and automation work in Vellum — one-shot and recurring schedules, heartbeats, watchers, playbooks, and proactive background work.
• Self-improving Skills: See how your assistant turns procedures it has carried out into reusable skills that you can review, update, and remove.
• Tools & Skills: Skills vs. tools in Vellum — atomic actions, capability bundles, built-in skills, and creating custom skills.
• The Workspace: The Vellum workspace — folder structure, IDENTITY.md, SOUL.md, USER.md, LOOKS.md, config.json, and skills directory.
• Web Search: How web search works in Vellum: managed vs bring your own, the four built-in providers (Provider Native, Perplexity, Brave, Tavily), the fallback chain, and how it's billed.
• Brave Search: How to connect Brave Search as your web search provider in Vellum: API key setup, provider configuration, and usage.
• Firecrawl: How to connect Firecrawl as your web search and web fetch provider in Vellum: API key setup, provider configuration, and usage.
• Perplexity: How to connect Perplexity as your web search provider in Vellum: API key setup, provider configuration, and usage.
• Tavily: How to connect Tavily as your web search provider in Vellum: API key setup, provider configuration, and usage.
• Pricing: How Vellum pricing works today: prepaid credits, pending usage, and how to add credits to your account.
• Privacy Notice: Vellum's privacy notice explaining how we collect, use, disclose and otherwise process your information.
• Vellum Prohibited Use Policy: Vellum's prohibited use policy — ensuring the Services are used safely, ethically, and in accordance with all applicable laws and regulations.
• Releases - Vellum Docs: Vellum release notes, versioning, update channels, and how to stay up to date with the latest features and fixes.
• Roadmap: What's coming to Vellum — shipping soon, up next, on the horizon, and what we're exploring.
• Skills Reference: Vellum skills reference — documentation for all available skills, capabilities, and configuration options.
• ACP: ACP skill for Vellum — delegates development tasks through the Agent Client Protocol.
• Amazon: Amazon skill for Vellum — shop, track orders, and manage your Amazon account through your assistant.
• App Builder: App Builder skill for Vellum — create interactive web apps with HTML, CSS, and JavaScript through conversation.
• Browser: Browser skill for Vellum — navigate web pages, interact with content, and extract information.
• ChatGPT Import: ChatGPT Import skill for Vellum — imports your conversation history from ChatGPT into Vellum.
• Computer Use: Computer Use skill for Vellum: control supported apps on macOS and Windows with screen observation, clicking, and typing.
• Contacts: Contacts skill for Vellum — manage contacts, communication channels, access control, and invite links.
• Document: Document skill for Vellum — create and edit long-form text in a dedicated rich-text editor with Markdown support.
• DoorDash: DoorDash skill for Vellum — order food, groceries, and convenience items through your assistant.
• Email (AgentMail): Email skill for Vellum — send, read, search, and manage email from your assistant's own address.
• Followups: Followups skill for Vellum — tracks messages awaiting responses across all communication channels.
• Gmail: Gmail skill for Vellum — full inbox management including archive, label, draft, send, and filter capabilities.
• Google Calendar: Google Calendar skill for Vellum — view events, create meetings, and check availability.
• Image Studio: Image Studio skill for Vellum — generate and edit images using AI through your assistant.
• Media Processing: Media Processing skill for Vellum — processes video, audio, and image files through a multi-phase AI pipeline.
• Messaging: Messaging skill for Vellum — send and receive messages across Slack, Gmail, and Telegram.
• Notifications: Notifications skill for Vellum — sends notifications through a unified routing system across connected channels.
• Phone Calls: Phone Calls skill for Vellum — make and receive calls via Twilio with real-time voice conversation and transcription.
• Playbooks: Playbooks skill for Vellum — trigger-action automation rules for handling incoming messages.
• Schedule: Schedule skill for Vellum — recurring and one-shot scheduled actions using cron syntax, RRULE patterns, or simple timestamps.
• Screen Watch: Screen Watch skill for Vellum — observe your screen at regular intervals with OCR for context-aware assistance.
• Skill Management: Skill Management skill for Vellum — creates and deletes custom managed skills to extend your assistant.
• Slack: Slack skill for Vellum — scan channels, summarize threads, manage reactions, and configure privacy-aware integration.
• Start the Day: Start the Day skill for Vellum — get a personalized daily briefing with weather, news, and tasks.
• Subagent: Subagent skill for Vellum — spawns autonomous background agents that work independently on tasks.
• Tasks: Tasks skill for Vellum — a two-layer task system with reusable templates and a prioritized work queue.
• Transcribe: Transcribe skill for Vellum — transcribes audio and video files using OpenAI Whisper or whisper.cpp.
• Watcher: Watcher skill for Vellum — polls external services for changes and notifies you when something happens.
• Weather: Weather skill for Vellum — get current conditions and forecasts for any location.
• Trust & Security: Vellum trust and security — privacy, data handling, permissions, and deployment best practices.
• Privacy & Data: Vellum privacy and data — what stays local, what leaves your device, and what Vellum never does.
• Security Best Practices: Security best practices for Vellum — sharing information, reviewing files, permissions, and credentials.
• The Permissions Model: Vellum's permissions model — risk tolerance tiers, sandbox boundary, trust rules, and the Allow/Deny permission prompt.
• Vellum Terms of Service: Vellum's terms of service — the agreement governing your use of the Vellum AI agent platform and services.