Vellum

Overview

Vellum is a personal AI assistant for people who want help acting across files, calendars, email, code, and a computer. It suits individuals willing to configure skills and permissions; its breadth is appealing, but it requires an internet connection and sends conversation context to Anthropic by default.

Key features

Vellum saves preferences and context in its workspace for use in later conversations, which can spare users from repeatedly explaining personal details. Custom skills let users extend the assistant by creating, installing, and managing capabilities, a useful degree of flexibility for people prepared to tailor their assistant rather than rely on a fixed set of functions.

OAuth integrations cover Discord, GitHub, Google services, HubSpot, Linear, Twitter (X), Asana, Notion, Microsoft Outlook, and Todoist. Vellum says tokens remain in a credential vault and are not exposed to the language model. Sensitive actions prompt for approval, while configurable trust rules allow standing permissions. Those controls help put boundaries around an assistant that can take action, though users still need to choose which permissions they are comfortable granting.

Vellum Cloud runs assistants in isolated environments with private storage and 24/7 availability. Conversation context goes to Anthropic by default; other listed model options are OpenAI, Google, OpenRouter, Fireworks, and local Ollama. Vellum says it does not use user data for model training, encrypts data end-to-end, and conducts continuous security audits. Optional usage analytics and crash diagnostics are off by default. These commitments may reassure privacy-conscious users, but the default routing is worth weighing before sharing sensitive context. Vellum also says the assistant cannot respond without internet access, even though local Ollama is an option.

Pricing

Vellum has a free plan and paid plans from $30 per month. Free starts with first tasks included and then uses pay-as-you-go credits, so it offers a way to begin without a monthly subscription but does not promise an ongoing included task allowance.

Mighty is the entry paid tier for an individual who wants a small cloud computer and included usage. Super and Ultra raise computing resources and storage; both add an assistant email and subdomain, making them more suited to users who need those resources or features. The free option avoids a recurring charge, but its pay-as-you-go credits differ from the included usage of paid tiers.

Platforms

Vellum supports Android, iOS, macOS, Windows, web, and a browser extension, as well as self-hosting. There is no shipped Linux desktop client; Linux users can instead use the web app, Chrome extension, or self-host the runtime. That range gives Linux users routes to access the assistant, but not a native desktop app. The internet requirement also rules it out for users who need an assistant that works offline.

Who it's for

Vellum is aimed at individuals seeking an assistant that can retain context and work across connected services, with approval prompts and trust rules for sensitive actions. It is less suitable for teams looking for a shared company tool, users who require offline operation, or anyone unwilling to route conversation context to a model provider.

Pros and cons

• Pro: Saved context and custom skills give individuals a way to build an assistant around recurring needs.
• Pro: A credential vault, approval prompts, and configurable trust rules provide controls for connected accounts and sensitive actions.
• Pro: Cloud plans include a hosted computer, storage, and usage; Vellum Cloud assistants run in isolated environments with private storage and 24/7 availability.
• Con: The assistant requires internet access, so it is not an option for offline use.
• Con: Anthropic receives conversation context by default, which may not suit users who want a different default data route.
• Con: Linux users must choose web, extension, or self-hosting rather than a shipped Linux desktop client.
• Con: Vellum is positioned as a personal assistant, not a shared team or company tool.

Alternatives

For work centered on prompt management, evaluation, or workflow tooling rather than a personal assistant, consider AI Prompt Management Software, AI LLM Evaluation Tools, AI Workflow Automation Software, Prompt Management Software, AI Prompt Generators, or AI Agent Platforms.

Promptfoo is a freemium option with a free Community plan that includes 10k red-team probes per month, evaluation features, and local or self-hosted use, making it a better fit for LLM evaluation and red-teaming than a personal assistant. PromptHelm has a free tier capped at 10,000 gateway calls per month, 10 prompts, and one seat; it is a more focused option for prompt and gateway management. Agenta offers a free Hobby plan with two team members, 5,000 agent runs per month, and 20 evaluations per month, so it better fits a small team working on agent evaluation. Braintrust offers a free Starter plan with 1 GB processed data, 10,000 scores, and 14-day retention, a fit for users focused on scoring and evaluation. Langfuse has a free Hobby plan with 50k units per month, 30 days of data access, and two users, suiting users who need usage and trace access for a small group. Maxim AI offers a free Developer plan for up to three seats, one workspace, and 10k logs per month, for teams that want a limited evaluation and logging start. MLflow GenAI Evaluation is a free open-source option with an evaluation API and UI for users who want that narrower evaluation tool. Prompt Manage has a free plan capped at 25 prompts and 10 runs per month, with a $20 per month Team plan for unlimited prompts; it suits prompt organization and runs rather than a general personal assistant.

Verdict

Choose Vellum if you want an individual AI assistant that remembers context, extends through custom skills, and can act across connected services with approval controls. Its strongest case is the combination of persistent context and practical permissions; look elsewhere if offline access, a native Linux desktop client, a shared team product, or a different default route for conversation context is essential.