Vellum

Summary

Vellum is an open-source personal AI assistant with long-term memory, its own identity, and tools for email, calendar, web browsing, files and computer control. It runs in Vellum Cloud or self-hosted, reachable from desktop, mobile, web, CLI and chat channels.

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

Captured 2026-09-08 · theagentsindex.com

From: Freethen $30/month
Free tier: YesBase plan, free forever: Small machine (1 vCPU, 2 GiB RAM), 6 GiB persistent storage, Vellum-managed model credentials, and pay-as-you-go credits with no monthly minimum. Paid inference, web search and image generation still consume credits bought in $10 increments.

Visit Vellum Compare Find alternatives

What it does

Writes code: YesThe assistant writes code and builds small apps itself, alongside roughly thirty bundled skills; the App Builder skill creates interactive web apps from conversation.Gmail management, Google Calendar, Slack integration, web browsing, computer control, phone calls, image generation, coding, app building, document…vellum.ai/docs/help/faq.md
Debugs and fixes: PartlyDebugging happens mostly by delegation: a read-only researcher subagent does root-cause analysis, and the ACP skill hands bug fixes to external coding agents you authenticate yourself.Self-contained tasks. Best for tasks like fixing a bug, refactoring a file, or writing tests.vellum.ai/docs/skills-reference/acp.md
Writes tests: PartlyTest writing is listed as work for delegated external coding agents through the ACP skill rather than as a first-party capability of the assistant's own coding tools.Self-contained tasks. Best for tasks like fixing a bug, refactoring a file, or writing tests.vellum.ai/docs/skills-reference/acp.md
Reviews code: NoDiff review is not shipped: PR review sits in the 'On the horizon' band of the vendor's own roadmap and appears in no skill page.Your assistant reviews pull requests with your standards, catches regressions, and flags the things you'd catch yourself.vellum.ai/docs/roadmap
Whole-codebase work: NoACP delegation is scoped to single self-contained tasks; coordinating agents across repositories is roadmap work, not a shipped capability.One assistant that coordinates multiple coding agents across your repos — planning, delegating, and shipping end-to-end.vellum.ai/docs/roadmap
Ships it: NoBuilt apps render in the workspace panel only; publishing or deploying what the assistant builds is listed as future work.Ship something your assistant built with a single share link — dashboards, tools, little apps — no deploy step required.vellum.ai/docs/roadmap
Done-for-you service: NoVellum sells self-serve software you run yourself, on its cloud or your own hardware, with no staffed build-and-run engagement offered anywhere on the site.Vellum assistant is open source. Deploy it on your own hardware, a VPS, or a private cloud. You control your data, your keys, and your uptime SLA. No…vellum.ai/pricing

How much it does unattended

Runs autonomously: PartlyHeartbeats, schedules, watchers and subagents run unattended in the background, but at the default risk tolerance anything that changes state still stops for your approval.Only what you allow it to. Vellum has a fine-grained permissions system built to protect your data. High-stakes actions like sending emails to new…vellum.ai
Multi-agent: YesThe assistant fans work out to several background subagents at once, each with a researcher, builder or advisor role, though subagents cannot themselves spawn subagents.Spawns autonomous background agents (subagents) that work independently while your main conversation keeps going. Your assistant can fan a single…vellum.ai/docs/skills-reference/subagen…
Agent permissions: YesFour risk-tolerance tiers from Strict to Full access, set separately for chat, scheduled tasks and API-triggered work, with deterministic classification outside the model sandbox.Your risk tolerance controls the threshold below which actions auto-approve without prompting. You can configure it in Settings > Permissions &…vellum.ai/docs/trust-security/the-permi…

Which models it runs on

Claude: YesAnthropic's Claude is the default provider for managed cloud assistants, and Vellum supplies the API credentials on the free Base plan.By default, data is routed to **Anthropic** (Claude) using Vellum's credentials.vellum.ai/docs/trust-security/privacy-a…
GPT: YesOpenAI models are selectable, and the shipped Quality, Balanced and Fast inference profiles are all built on GPT models out of the box.If you connect your own API key, you can switch to **OpenAI**, **Google** (Gemini), **OpenRouter**, or **Fireworks**.vellum.ai/docs/trust-security/privacy-a…
Open models: YesOpen-weight models are supported, including a fully local Ollama endpoint so inference never leaves your machine; the Budget profile ships on an open model.you can switch between models including OpenAI, Anthropic, Gemini, or open weights via Ollama at any timevellum.ai
Model choice: YesYou pick the model per workspace, per conversation and per action type through inference profiles and action overrides, with custom profiles for providers not built in.Open **Settings → Inference Profiles** and select a different active profile from the dropdown. The change applies workspace-wide immediately — no…vellum.ai/docs/key-concepts/model-profi…

Where you use it

In your editor: NoThe vendor's enumeration of first-party surfaces lists browser, mobile, desktop and CLI only; there is no editor or IDE plugin, only a Chrome extension for browsing.On Vellum Cloud, you can reach your assistant from any modern browser at [vellum.ai](https://vellum.ai), the iPhone and iPad app, the Android app,…vellum.ai/docs/help/faq.md
On the command line: YesA first-class `vellum` CLI sends messages, tails the event stream and manages processes, and is also how self-hosted runtimes are started and inspected.The `vellum` CLI is the highest-level interface. It reads your local credentials and routes commands to the running daemon.vellum.ai/docs/developer-guide/get-star…
In your pipeline: NoNo pull-request or CI surface in the channel/interface list or the 29-skill catalog; PR review and webhooks are roadmap items.Wire any service into your assistant with a simple webhook. If it can POST, your assistant can react to it.vellum.ai/docs/roadmap
In a browser: YesA hosted web app is one of the supported surfaces and the recommended route on Linux, where no desktop client ships.The web app and Chrome extension work on Linux. There is no Linux desktop client to install.vellum.ai/docs/help/faq.md

Whose machine it runs on

Self-hosted: YesThe runtime can run on your Mac, a GCP VM in your own project, or any Linux host you can SSH into, with no platform or compute fees.Run the assistant on infrastructure you control. This is useful when you need the assistant to stay running independently of your local machine, when…vellum.ai/docs/environments.md
Open source: YesThe runtime, gateway, CLI, clients and skill catalog are published as an MIT-licensed monorepo on GitHub, which the vendor cites as an auditability argument.Open-source (MIT) monorepo with the assistant runtime, CLI, clients, gateway, and skill catalog.vellum.ai/llms.txt

What it costs to run

How it meters: YesTwo meters at once: a monthly subscription for machine, storage and an included usage bundle, plus pay-as-you-go credits where one credit equals one US dollar of provider cost.Credits are how you pay for what your assistant does, things like AI responses, web searches, and image generation. $1 = 1 credit, and you only pay…vellum.ai/pricing
Free tier: YesA standing free Base plan gives a small machine, 6 GiB storage and Vellum-managed model credentials, with no sales call and no card needed to start.Free, forever. Includes everything you need to run an assistant:vellum.ai/docs/pricing.md
API access: PartlyA versioned REST API and SSE stream exist with JWT auth, but the vendor states the HTTP API is gated behind the gateway on Vellum Cloud, so raw API work needs a local install.The HTTP API is gated behind the platform gateway in cloud mode, not exposed directly, so for raw API development you'll want a local install.vellum.ai/docs/developer-guide/get-star…
MCP server: PartlyVellum is an MCP client: it connects to MCP servers added in settings or shipped by plugins, but the docs describe no MCP server through which another agent could query the assistant.A plugin declares MCP servers in a root mcp.json. The assistant connects them on install and registers their tools alongside workspace-configured MCP…vellum.ai/docs/llms.txt
Bring your own key: YesYou can bring your own provider keys and switch away from Vellum-managed Anthropic access, or point inference at a local Ollama instance entirely.If you configure your own connection to a third-party AI provider through Vellum, then your data is sent directly to that provider using your…vellum.ai/docs/privacy-policy.md

Buying it for a team

A company can buy it: PartlyA company can buy it and the terms bind an Organization, but every published plan is a single-assistant subscription: no team plan, seat management or admin console.Your company gets a GTM team. Starting at $30/month.vellum.ai/business
Seat model: NoNothing is priced per seat: the vendor's own breakdown of a monthly total is platform fee plus machine tier plus storage tier plus an optional usage bundle.Your monthly total is the sum of: - **Platform fee** ($10/mo): custom subdomain, static IP, priority support - **Machine tier**: $35, $60, or $125/mo…vellum.ai/docs/pricing.md
Pooled budget: NoEach account holds its own credit balance, capped at 1,000 credits, and the terms forbid moving credits between users, accounts or organizations.Vellum Credits are non-transferable between users, accounts, or Organizations.vellum.ai/docs/vellum-terms-of-use
Admin controls: PartlyThe owner can constrain the agent tightly through risk tolerances, trust rules and per-context thresholds, but the vendor states there is no admin panel above the individual user.Your assistant is yours. Vellum doesn't have a dashboard, an admin panel, or any way for your employer to see your usage.vellum.ai/docs/help/faq.md
Audit log: NoNo administrative audit trail: the vendor states there is no admin panel. The account owner gets a usage breakdown by day, model and action, and conversation history, which is not the same claim.Vellum doesn't have a dashboard, an admin panel, or any way for your employer to see your usage.vellum.ai/docs/help/faq
Single sign-on: PartlyConsumer social sign-in through Apple or Google is documented; no SAML or enterprise OIDC federation appears anywhere in the docs, pricing or privacy notice.if you create or log into your Vellum account through a third-party platform (such as Apple or Google), we may have access to certain information…vellum.ai/docs/privacy-policy.md

What happens to your code

Opt out of training: YesConversation-content collection is an explicit opt-in/opt-out toggle, telemetry is off by default, and the vendor states data is not used to train models either side.use it to improve the quality and accuracy of our Services, including evaluating and improving the systems that power Vellum, but do not use it to…vellum.ai/docs/privacy-policy.md
Data residency: PartlyNo region picker for the managed cloud, and the vendor warns of US-based storage and cross-border transfers; the jurisdiction is yours only if you self-host on your own machine or VM.Vellum is headquartered in the United States and we have operations and vendors in the United States and other countries. Therefore, we and our…vellum.ai/docs/privacy-policy.md
Getting out: YesThe workspace is exportable at any time and deletable on request, and self-hosted installs can be archived to a tarball from the CLI before removal.You can export your workspace at any time, or request full deletion through the account settings or support.vellum.ai/docs/help/faq.md

Checked 2026-09-27 · How we check

Filed under

APIAutonomousBrowser useClaudeFree tierGPTMulti-agentOpen modelsOpen sourceSelf-hosted

Sources and updates

Sources on vellum.ai35

Coverage elsewhere3

Updates6

2026-09-27Capabilities · Free tier · Terms
2026-09-03Capabilities · Free tier
2026-09-02Terms
2026-08-31Capabilities · Free tier · Links · Price · Terms
2026-08-30Capabilities · Links · Terms
2026-08-19Summary · Tagline