Vellum Review, Pricing & Alternatives (2026)

Vellum is an open-source personal AI assistant built by Vellum AI for business and personal work. It remembers preferences, maintains an identity, and acts across connected tools instead of stopping at generated text. The company introduced its Personal Intelligence positioning on May 7, 2026, and announced $25 million raised from Dharmesh Shah, Arash Ferdowsi, Rebel Fund, and Y Combinator. The project repository was created on February 7, 2026, uses the MIT license, and reports 1,450 stars and 219 forks. Supported surfaces include macOS, Windows, iOS, web, voice, email, Telegram, Slack, and Twilio. Local installations keep data on the user’s machine, while Cloud provides an isolated encrypted container. Pricing starts with a free Base plan; Pro packages cost $30, $100, or $200 monthly. Vellum suits users who want an assistant they can self-host, customize, and supervise.

A typical workflow begins with conversation, then draws on memory, skills, scheduled checks, and approval rules. Vellum can read, draft, and prioritize email, manage calendars, browse the web, edit files, create documents, handle GitHub issues, summarize Slack activity, and run recurring tasks. Its repository lists eight memory types, per-user and per-channel isolation, hybrid retrieval, local embeddings by default, and hourly checks for unfinished or soon-due work. OAuth covers Slack, Notion, Google, HubSpot, Linear, Discord, Twitter, Telegram, and Twilio. Plugins add tools and prompt sections through SKILL.md and TOOLS.json files, while 60-plus built-in skills cover services including Google Workspace, Slack, Notion, and the filesystem. Writing files, shell commands, messages, deletion, and computer use can require approval, with trust rules granting narrower or broader permission. It suits users who want an assistant that acts across their tools, while developers may pick OpenClaw.

Vellum’s memory story contains a meaningful architectural split. Its May 2026 product account describes four always-loaded files: essentials.md for costly-to-forget facts, threads.md for open commitments, recent.md for the last day or two, and buffer.md for unfiled memories. A consolidation pass runs every four hours, while corrections are fast-tracked into essentials.md. Behind them sits a graph whose pages use BM25 and dense embeddings, PCA, summaries, and spreading activation to retrieve related context. The repository presents a different implementation: eight memory types, hybrid dense and sparse retrieval, source attribution, deduplication, per-user and per-channel isolation, and local ONNX embeddings with cloud fallback. Model choice is broad, spanning Anthropic, OpenAI, Gemini, Fireworks, OpenRouter, MiniMax, Atlas Cloud, compatible endpoints, and Ollama. The recent releases suit users who value frequent features, while cautious teams should pick tool.

Vellum’s free Base tier provides a Small machine with 1 vCPU and 2 GiB RAM, 6 GiB of persistent storage, managed LLM credentials, and pay-as-you-go credits without a monthly minimum. Pro packages add recurring usage allowances. Mighty costs $30 monthly for Small compute, 10 GiB storage, and Mighty Usage, but excludes the $10 platform fee. Super costs $100 for Medium compute at 2.5 vCPU and 5 GiB RAM, 30 GiB storage, Super Usage, a custom subdomain, static IP, and priority support. Ultra costs $200 for Large compute at 4 vCPU and 8 GiB RAM, 60 GiB storage, Ultra Usage, a custom subdomain, static IP, and priority support. Custom configurations start at $50 monthly. One credit costs $1 and covers inference, web search, image creation, and paid managed APIs, making Base suitable for light users and Pro preferable for heavier workloads.

Vellum’s main limitation is operational and financial complexity. Hosted use combines machine capacity, storage, monthly usage bundles, prepaid credits, and model-provider charges. Base is free with a 1 vCPU, 2 GiB machine and 6 GiB storage, while Pro packages cost $30, $100, or $200 monthly; included usage expires each billing cycle, and purchased credits expire after twelve months. Background heartbeats and memory consolidation can consume credits even when no chat is active. Self-hosting removes platform and computer-tier fees, but users assume responsibility for uptime, updates, backups, and model inference. ChatGPT and Claude remain more familiar conversation tools, whereas Vellum maintains memory, runs scheduled work, and acts across channels. OpenClaw is a closer comparison for developers, offering greater autonomy; Vellum instead provides guided onboarding and progressive approvals. People wanting persistent personal context and controlled recurring work should pick Vellum.