Product Roadmap

Your assistant doesn't go dark when your laptop closes. Move seamlessly between local and cloud — same memory, same identity, same conversations.

Wire any service into your assistant with a simple webhook. If it can POST, your assistant can react to it.

A native Apple-ecosystem channel for your assistant — texts, read receipts, the works.

I want my assistant to join my calls to take notes and have the context of the things Im working on. Ideally it can start tasks discussed in meetings before I tell my assistant to get on them lol.

Your assistant can talk to other assistants — coordinating schedules, sharing context, negotiating on your behalf.

Extend the subagents primitive to enable the user to define their own agent personas that can be used for the root conversation and not just the subagents themselves

A shortcut making a floating mini chat area to chat with assistant, with the additional ability to "pin" it so that if I click on another application it doesn't delete

incognito mode where it won't make memories from the conversation and you can choose if memories to in or not

Your assistant can take and make phone calls on your behalf — screening, scheduling, and handling the calls you don't want to.

The small recurring stuff — birthdays, renewals, reservations, errands — handled without a spreadsheet.

A work assistant and a personal assistant. Or one per project. Separate memories, separate identities, same you.

Summon your assistant inside any app on your system — summarize a doc, reply to a message, translate a selection — without switching windows.

Make it so you can freely download TTS models on your computer and the assistant could use that. Possibly with some first class integration with HuggingFace. This would be great for image generation, general language, embedding, and any other modality for models.

Add the ability to edit and resend the previously sent message to fix errors / typos without needing to correct with a new message

Would love to see the Vellum assistant available as a self-hosted option on Linux. Currently the assistant is macOS-only for local hosting, which locks out a large portion of developers and power users who run Linux as their primary OS or on home servers. Use cases include: runni…

I’d love to see a browser agent that can continue multi-step web workflows across sessions: downloading ADP paystubs, uploading YouTube videos, filling forms, organizing files, and resuming interrupted tasks. Ideally it would work with memory, permissions, and Vellum Doctor so us…

Your assistant on your wrist — glanceable, taptic, and ambient.

Pull signals from the wearables you already have and turn them into useful nudges, not a dashboard you have to check.

A community hub for teaching your assistant new tricks. Pull in skills others have built; publish your own.

The macOS desktop client registers a GLOBAL shortcut on CmdOrCtrl+Shift+G by default (GLOBAL_SHORTCUT_DEFAULTS.globalHotkey). A global registration outranks every app's own shortcuts, so in Adobe Illustrator, Shift+Cmd+G (ungroup — a core, constant-use command) silently does noth…

On the macOS desktop app, multi-step turns that finished normally still render as a collapsed 'Thinking... N steps' chip, and earlier messages get moved under an 'Earlier activity' divider while the user is still reading them. The full text exists but sits behind folds the user d…

Assistants should recognize different voices during a voice conversation. Users should be able to create a voice profile for a given contact either before or during a voice session, and their assistant should use that to know who is speaking.

Add an integration for https://www.agentmail.to/, ideally as a plugin/skill

Select text ANYWHERE on your screen, a button appears displaying multiple actions like correcting spelling and grammar (the actions shouldn't have any personalization), or you can type your own custom action.

Give my assistant its own phone number for sending and receiving text messages