Vellum

Vellum is an AI assistant and AI development platform that helps users build, deploy, and operate AI-powered applications and assistants. The product originally launched in 2023 as an AI development platform focused on helping companies evaluate, manage, and experiment with large language models and prompts. Its current product has expanded into a personal AI assistant that can work across files, applications, communication channels, calendars, websites, and other services.

The current Vellum assistant is designed to go beyond conventional chatbot interactions. Rather than only generating responses, it can perform actions on a user’s behalf through connected tools and integrations. Vellum describes the assistant as having its own identity, persistent memory, filesystem, accounts, and ability to interact with external services. It can browse the web, work with files, send emails, manage calendars, build applications, control a Mac, and communicate through channels such as Slack and Telegram.

Persistent memory is a central part of the current product. Information, preferences, and relevant context can be retained across conversations so that users do not need to repeatedly provide the same instructions. Vellum’s documentation describes a memory and context system that allows assistants to recall information from previous interactions and use it when relevant to future tasks.

Vellum also provides a workspace where the assistant can store files, notes, conversation history, skills, and other information. This gives the assistant an environment in which it can work on tasks rather than simply responding to individual prompts.

The platform includes a skills system that gives the assistant access to specific capabilities. Built-in skills cover areas such as Gmail, Google Calendar, Slack, messaging, phone calls, browser automation, coding, image generation, task management, and document-related work. Users can also create or install additional skills to extend what the assistant can do.

Vellum supports browser automation, allowing the assistant to navigate websites, interact with page elements, extract information, and fill forms. This makes the assistant useful for tasks that cannot be completed solely through a conventional API integration.

The platform also supports computer control on macOS. Through macOS accessibility APIs, the assistant can interact with applications and perform actions on the user’s computer. The desktop application also provides voice input and connects to the same assistant available through Vellum Cloud.

Vellum can be accessed across several surfaces. Users can interact with the assistant through a web browser, macOS desktop application, iOS application, Slack, Telegram, email, and phone calls. This allows the same assistant, memory, tools, and workspace to be available across different environments.

Communication capabilities are another major part of the platform. Vellum can connect to Gmail and manage inboxes, draft and send messages, search email, archive messages, and perform other supported email actions. It can also use its own email identity to send and receive messages.

Google Calendar integration allows the assistant to view schedules, check availability, create events, modify appointments, and manage calendars. Users can ask the assistant to identify available time slots or make scheduling changes through natural-language instructions.

Slack integration allows Vellum to read and search channels, summarize conversations, send messages, manage reactions, and interact with workspace communication. Telegram provides another messaging channel through which users can communicate with their assistant.

Vellum also supports integrations with services including GitHub, Google Workspace, HubSpot, Linear, Twitter/X, Asana, Notion, Microsoft Outlook, and Todoist. These connections are generally established through OAuth2, with access credentials stored in a secure credential vault.

The assistant can work with business workflows as well as personal productivity tasks. Vellum highlights examples including reviewing sales pipelines, updating CRM records, enriching and scoring leads, preparing competitor briefs, scheduling quarterly business reviews, flagging contract changes, producing meeting notes, and identifying churn risks.

Vellum also supports application development. Its assistant can write code and build applications, while its developer platform provides tools for constructing more structured AI workflows. This makes the product relevant to both end users who want an AI assistant and technical users who want to develop AI-powered systems.

The developer side of Vellum provides a Workflows SDK for defining and executing complex AI systems as graphs. Developers can use workflows to construct more structured AI applications rather than relying exclusively on free-form assistant interactions.

Vellum’s development environment also includes capabilities for experimentation, evaluation, deployment, monitoring, and collaboration. These capabilities were central to the product’s original AI-development-platform positioning and remain part of the broader Vellum ecosystem.

The platform provides official client libraries for Python, Node/TypeScript, and Go. Its API documentation provides HTTP endpoints for interacting programmatically with Vellum, and the platform can be deployed in cloud-hosted or self-hosted/VPC environments.

Python is particularly important to the developer ecosystem. The Vellum Workflows SDK requires Python 3.9 or higher and allows developers to define workflows programmatically.

Vellum also supports self-hosting. The current product documentation states that Vellum Assistant is open source and can be deployed on a user’s own hardware, VPS, or private cloud. Self-hosting provides greater control over data, API keys, and uptime.

The platform takes a permission-based approach to sensitive actions. Vellum states that actions such as sending emails, accessing files outside the assistant’s workspace, controlling the user’s screen, making purchases, or connecting new services can require explicit approval. Users can configure whether approval is required once, temporarily, or through a standing rule.

Vellum also provides scheduling and automation capabilities. Assistants can run recurring or one-time scheduled tasks, use routines and heartbeats, and continue performing certain work while the user is away. This allows the assistant to operate more like an ongoing digital worker rather than a system that only responds when directly prompted.

The platform supports multiple AI models and model profiles, allowing users to determine which language model handles particular jobs. This gives users greater flexibility when balancing capability, performance, and cost.

Vellum’s pricing model is also different from many conventional AI subscriptions. The company passes model-provider costs through at cost without adding a markup to token usage. Users can consume included monthly usage on paid plans and then use prepaid Vellum Credits for additional usage. One Vellum Credit equals one U.S. dollar.

The current plans include a Free Base plan, followed by paid Mighty, Super, Ultra, and Custom configurations. The Base plan is free and includes a small computer, persistent storage, and pay-as-you-go credits. Paid plans increase computing capacity, storage, and included monthly usage.

The Mighty plan costs $30 per month, Super costs $100 per month, and Ultra costs $200 per month. Mighty includes a small computer and 10 GiB of storage, Super includes a medium computer and 30 GiB, while Ultra includes a large computer and 60 GiB.

A Custom configuration is also available for users who want to select individual machine, storage, and usage components. The documentation currently lists a $10 monthly platform fee for applicable Custom configurations, with additional charges based on compute, storage, and optional usage bundles.

Vellum is therefore suited to users and teams looking for an AI assistant that can remember context, interact with applications, automate tasks, communicate across services, browse the web, work with files, write code, and perform scheduled activities.

Its combination of personal-assistant functionality and developer tooling makes it broader than a standard chatbot. Users can interact with it conversationally, while developers can use APIs, SDKs, workflows, integrations, and self-hosting options to create more advanced AI systems.

Pricing

Vellum currently offers a Free Base plan and paid plans starting at $30/month. The paid tiers are Mighty at $30/month, Super at $100/month, and Ultra at $200/month. A Custom configuration is also available.

Vellum does not add a markup to model-provider token costs. Instead, it passes those costs through at cost and uses Vellum Credits for certain paid features such as AI inference, web search, image creation, and selected third-party APIs.

Pricing Model

Paid

Pricing Note

Vellum has a Free Base plan and paid plans starting at $30/month. Current paid tiers are Mighty ($30/month), Super ($100/month), and Ultra ($200/month), with a Custom configuration also available. Usage beyond included allowances is handled through Vellum Credits, with $1 = 1 credit.

Review

Vellum is an interesting AI assistant because it focuses on taking actions rather than simply generating answers. Its persistent memory, integrations, browser automation, computer control, scheduling, and communication capabilities make it capable of handling more involved workflows than a conventional conversational AI.

Its strongest feature is the combination of memory and tool access. The assistant can retain useful context while using connected services such as Gmail, Google Calendar, Slack, GitHub, Notion, HubSpot, and Asana.

The broad range of access points is another advantage. Users can reach the same assistant through the web, macOS, iOS, Slack, Telegram, email, and phone calls.

Vellum’s free entry tier is also attractive because users can start without a subscription. Paid plans provide greater compute, storage, and included usage, while the credit system allows additional usage when required.

Self-hosting is another significant advantage for technically capable users who want greater control over their infrastructure and data.

The main limitation is that the product’s broad permissions and automation capabilities require careful configuration. Users need to understand which services they connect and which actions they authorize. Vellum’s approval system helps address this by requiring confirmation for sensitive operations where configured.

Another consideration is usage-based cost. Although the Base plan is free, AI operations can consume credits, and paid plans can incur additional usage charges once included allowances are exhausted. This makes monitoring usage important for users running assistants frequently or automating background tasks.

For users who want more than a chatbot and need an assistant that can interact with their digital environment, Vellum offers a broad combination of memory, automation, integrations, coding, browser interaction, and multi-device access.