Vellum launches its self-hostable personal AI and raises $25 million

Vellum launches its self-hostable personal AI assistant and raises $25 million from Y Combinator and tech founders to deploy open-source agent workflows.

Vellum.ai, a platform previously focused on enterprise AI agents (Y Combinator W23, $20 million Series A in 2025), is expanding its offering with Vellum: Your Personal Intelligence, a self-hostable, open-source personal AI assistant under an MIT license. The move is accompanied by a new $25 million funding round, with participation from Y Combinator, Leaders Fund, Dharmesh Shah (HubSpot co-founder), and Arash Ferdowsi (Dropbox co-founder).

Its usage is based on a three-step workflow named Hatch, Teach, Rely. Users hatch their assistant by giving it a name, train it through their corrections and habits, then delegate tasks to it as trust builds. The autonomy level is adjustable, from strict mode to full access, with each sensitive request being justified by the agent.

What distinguishes Vellum from simple conversation logging lies in its memory architecture. The system combines short-term working memory, procedural memory, and automatic consolidation every four hours that organizes elements into a knowledge graph and structured files (essentials, ongoing threads, recent events). The assistant can be connected to Slack, a calendar, or an email inbox to summarize threads, identify actions to take, and act autonomously on tasks such as rescheduling a meeting or preparing a schedule.

In terms of capabilities, Vellum is announced in beta for voice meeting participation, and offers end-to-end generation of short animated videos in under five minutes. The code is available on GitHub for local deployment, and the 𝕏 account @ashvellum, operated by Akash Sharma's personal assistant, provides a public overview of its operation.