# Vellum — A personal AI assistant with persistent memory, built-in skills, and phone/voice access, hosted in the cloud or self-hosted.

> Source: The Agents Index — https://theagentsindex.com/vellum-ai (structured, researched, re-verified)
> Facts last verified: 2026-08-19

Vellum is a personal AI assistant that runs in a hosted Vellum Cloud environment or on a user's own hardware. It carries its own identity, including its own email and accounts, and stays reachable across web, macOS, iOS, Slack, Telegram, and a terminal. It targets individuals and small teams who want one assistant that remembers context over time and can act across their tools rather than a single-purpose chatbot.

| Fact | Value |
| --- | --- |
| Website | https://vellum.ai |
| Pricing | Free Base plan, with paid packages starting at $30/mo (Mighty) |
| API | Yes |
| Not for | Teams needing centralized, seat-based access control or audited multi-user administration, since Vellum prices and operates as a single personal assistant package rather than a team management platform |

## Pricing

| Tier | Price |
| --- | --- |
| Base | $0 / month |
| Mighty | $30 / month |
| Super | $100 / month |
| Ultra | $200 / month |
| Custom | from $50 / month |

## Verdict

Vellum positions itself as a personal AI assistant rather than a narrow chatbot: persistent memory, a broad built-in skill set, and reach across web, mobile, Slack, Telegram, and phone calls set it apart from single-channel tools. The permission-gated action model and open-source self-hosting option are genuine differentiators for users wary of handing over unsupervised account access. The tradeoff is a pricing structure with several moving parts (machine tier, storage, platform fee, credit bundles) and third-party sources that disagree on what the product even charges, which makes it hard to budget confidently before signing up. Worth trying on the free Base plan first to see whether the skill set and permission model fit before committing to a paid package.

## How it works

1. Sign up for the free Base plan (no credit card required) and set up the assistant with its own identity.
2. Connect the built-in skills you want, such as Gmail, Google Calendar, Slack, or Linear, so the assistant can act on those accounts.
3. Chat, talk, or message the assistant from Web, macOS, iOS, Slack, Telegram, or a terminal, and approve or deny any sensitive action it requests.
4. Upgrade to a Mighty, Super, Ultra, or Custom package, or self-host, once you need more machine power, storage, or credits.

## Who it's for

- Individuals who want one AI assistant that remembers them across email, calendar, and chat instead of juggling separate tools
- People comfortable approving sensitive actions manually in exchange for an assistant that can act across accounts
- Technical users who want the option to self-host an open-source assistant rather than rely solely on a vendor's cloud

## Strengths and weaknesses

- ✓ Persistent memory and a wide built-in skill set (email, calendar, browsing, phone calls, coding) cover many personal tasks from a single assistant.
- ✓ Explicit Allow/Deny prompts with risk badges on sensitive actions give users a visible checkpoint before the assistant touches files, accounts, or the screen.
- ✓ Open-source self-hosting removes platform fees and lets users keep workspace data on their own hardware instead of the vendor's cloud.
- ✗ Pricing is layered across preset packages, a Custom configuration with separate machine, storage, and credit-bundle costs, and a $10/mo platform fee that applies only to some tiers, which makes total cost harder to estimate upfront.
- ✗ Independent write-ups of Vellum's pricing disagree on the plan names, prices, and even the product's shape (one third-party source describes prompt/workflow execution limits and a $500/mo Pro tier that don't match the vendor's current Base/Mighty/Super/Ultra structure), suggesting the offering has changed and stale information is circulating.
- ✗ Granting phone-calling, screen-control, and browser-extension access to a persistent-memory assistant raises a larger trust surface than a conventional chatbot, even with permission prompts in place.
- ⚠ Some capabilities, like outbound phone calls and computer control, depend on permission approvals working correctly each time; the vendor's own site is the source for most feature claims, and independent, hands-on reviews of these specific features are not available yet.

## Key features

- **Personal Knowledge Base** — Persistent memory that retains user facts, preferences, and conversation context across sessions rather than resetting each chat.
- **Built-in skill library** — Ships with skills for Gmail, Google Calendar, Slack, web browsing, computer control, phone calls, image generation, coding, app building, and task management, plus a growing library of installable or user-built skills.
- **Multi-channel presence** — Reachable on Web, macOS, iOS, Slack, Telegram, and a terminal, with a browser extension that reads open tabs and fills forms on the page you're viewing.
- **Permission-gated actions** — Sensitive actions like sending email, touching files outside its workspace, or running shell commands require an explicit Allow/Deny prompt with a risk badge before executing.
- **Voice and phone calling** — Supports push-to-talk or always-listening voice interaction, and can place outbound phone calls on the user's behalf using its own voice.
- **Self-hosting** — The assistant is open source and can run on a user's own hardware, a VPS, or a private cloud, with no platform fee or machine-tier charges in that mode.

## Use cases

- **Inbox and calendar handling** — A user connects Gmail and Google Calendar so the assistant can triage messages and manage scheduling using its built-in skills.
- **Ordering food hands-free** — A user asks the assistant to place a food order through its DoorDash integration without opening the app themselves.
- **Ticket and project updates** — A user has the assistant create Linear tickets, assign owners, and move ticket states from a conversational request.
- **Quick interactive tools** — A user asks the assistant to build a small interactive web app directly inside the conversation instead of opening a code editor.
- **Voice-driven tasks on the go** — A user switches to always-listening voice mode and has the assistant place an outbound phone call on their behalf.

## Integrations

Gmail · Google Calendar · Slack · X (Twitter) · Linear · DoorDash · Telegram

## Sources

- https://vellum.ai
- https://vellum.ai/docs/pricing
- https://vellum.ai/pricing
- https://vellum.ai/docs
- https://www.zenml.io/blog/vellum-ai-pricing
- https://hypertools.so/tool/vellum
- https://www.youtube.com/watch?v=ilKGyzGRhas
- https://www.eesel.ai/blog/vellum-ai-pricing