Dashy AI · In development · October

Your ideas. A place to act.

An Agent, personal memory, and help inside the widgets you already use. We are building AI into Dashy so useful answers become part of your day.

Try the original Dashy today4.710,000+ users

Interactive examples of the experience we are building, not live AI runs. Scroll to explore, or read the feature guide below.

Ask for a week, not a widget

“Get me through finals week.” Then close the chat.

One sentence to the assistant you already use: a board, a timer, what is due, the exam dates, and a room dimmed to match.

Assistant connected to Dashy

Ask anything

For whatever you are actually doing

“I’m running the marathon.” Race day, in the tab.

A countdown to the start line, the morning weather, your streaks, and a long-run card nobody had written before.

Assistant connected to Dashy

Ask anything

Not only for work

“We’re renovating the kitchen.” All of it, one screen.

The list, the measurements you keep losing, the day the worktop lands, and what is left of the budget. Tap a cost; the number moves.

Assistant connected to Dashy

Ask anything

The widget that does not exist yet

Describe a card. Get the card.

When the widget has never been written, your assistant writes it: a draft in a sandbox, added only when you say yes. The AI features are part of the paid plan.

New widget

Custom

Describe it

One Dashy. AI throughout.

From an idea to something you can use.

Dashy AI is being built into the whole dashboard: an Agent that helps set things up, remembers project context, and works with your tasks, notes, pages, and widgets. The aim is to help you act on information and pick up where you stopped, with a clear off switch.

The AI package

Useful on its own. Better together.

Local workflows tested; cloud rollout pending

Dashy Agent

Describe what you need, review a setup, and work with the actual dashboard objects. Tasks and notes shown in a conversation stay connected to the same saved items in your dashboard. Comparison tables in those notes can be edited in the conversation or in Notes. Open Resume from the dashboard to return to saved work. The Agent brings together browser actions, web search, files, and computation.

“Help me set up a dashboard for my product launch, with a task list, notes, and useful links.”

How it works and current limits

The current reviewed setup flow works with a selected dashboard. Page capture lets you choose a Tasks or Reading List destination and preserves its source. Resume is a standalone dashboard action that opens unfinished tasks, selected notes, and unread pages without a model call. Search saved work finds specific items in that dashboard, including completed tasks, and works with AI off. Web search in chat is designed to return source links. Files can supply selected text or images, and supported computation runs in a bounded sandbox.

Web search, file input, reasoning, and tools must be enabled and verified for the chosen route. Local saved-work search matches words in items attached to the selected dashboard’s supported widgets; it does not search every stored document. A source link is not a guarantee that an answer is correct. Voice is still gated while provider, audio, and cost checks are completed. The animated examples on this page illustrate the intended experience; they are not live Agent runs.

Account and sync flows tested locally; release pending

Personal AI memory

Keep useful decisions and preferences with the projects they belong to. Find saved context and return to the original source instead of repeatedly explaining your work.

“Remember that this project should use short, plain-language copy.”

How it works and current limits

Memory is separate from a chat transcript. In the development build, open Settings → AI → Personal memory to review, correct, or remove saved decisions and preferences, including with AI off. Local checks across two signed-in Chrome installations cover synced corrections, offline deletion after restart, and keeping an excluded project’s memories local. Dashboard retrieval returns references to supported saved objects in the selected dashboard.

This is not an automatic index of every page you visit or every file on your device. Current dashboard retrieval covers widget-bound Tasks, selected Notes, and Reading List entries. The local sync checks use two installations of the same development build; they do not establish compatibility with older extensions, every device, or the released service.

Integration and release checks in progress

Connect Claude and ChatGPT through MCP

Connect Claude or ChatGPT to supported Dashy tools through the Model Context Protocol (MCP). This lets an authorized assistant work with dashboard widgets, tasks and notes, rather than merely opening a chat shortcut. Compatible external tools can also connect to Dashy Agent.

Ask Claude or ChatGPT: “Add a task list for my launch and put it beside my notes.”

How it works and current limits

Connections require the appropriate account, connector configuration, and permissions. Access can differ by app and tool. Dashboard actions, saved-content access, and external tool calls are separate capabilities rather than one unrestricted permission.

MCP support does not mean every assistant, server, authentication flow, or tool combination has been verified. Use the supported setup shown in your installed version. Disconnecting a tool does not undo actions it has already completed.

Reviewed installation and recovery tested locally

AI-generated custom widgets

Describe a small tool you wish your dashboard had. Preview the generated widget, choose where it belongs, and keep using it alongside the built-in widgets.

“Make a card that counts the practice papers I have finished.”

How it works and current limits

Preview generated widgets before adding them; they run inside a sandbox. Saved designs can be revisited from history. Installation recovery keeps the same widget identity when a save confirmation is interrupted, avoiding duplicate additions on retry.

Generated code needs review. Sandboxing limits access; it does not prove every widget is correct. Network access depends on declared permissions. Persisting a widget’s own interactive data requires its supported storage interface; an arbitrary HTML input is not automatically synced.

Generation and apply flow under verification

Custom wallpapers

Describe a mood or a scene, preview a wallpaper, and choose whether to make it part of your dashboard.

“A quiet mountain lake, dark enough for readable white widgets.”

How it works and current limits

Wallpaper creation is a separate generation action with a preview and explicit apply step. Saved dashboard content remains usable when AI is off.

Generation depends on a configured image service and its usage limits. Preview quality, readable contrast, storage, and account behavior need checking before release. A wallpaper generation is not included simply because a chat model supports images.

Notes and capture flows tested locally

Help inside your widgets

Improve writing, summarize, translate, and extract tasks where the information already lives. AI assistance belongs in ordinary widgets rather than a separate collection of AI widgets.

Turn meeting notes into a short summary and a task list, then review what gets saved.

How it works and current limits

Notes assistance offers a result to review before replacing or appending text. Supported edits have version checks and undo where available. Captured pages and tasks can receive an optional destination suggestion; accepting it selects a destination, and Save remains a separate action.

Not every widget has contextual assistance yet. Native assistance depends on available browser capabilities and languages; unsupported local translations do not fall back to a paid model. Unsupported devices skip automatic suggestions. Changes made elsewhere can invalidate an older draft or undo action.

Manual reuse and local controls tested; release pending

Routines and background assistance

Save a useful workflow so you can return to it. Let small, bounded background jobs handle things like a chat title, a recap, or a capture destination suggestion.

Save a “Review my launch plan” instruction, then review and send it in a fresh chat.

How it works and current limits

Saved routines keep reusable instructions. Open one in a new chat, review it, and press Send when ready. Automatic local helpers reuse valid results and avoid repeating the same work. Background assistance has its own pause control, separate from the master AI switch.

The current passive helpers use already-available on-device AI, with no paid fallback or automatic model download. Saving a routine does not send its instructions. A recap does not mean work happened while you were away.

Model choice

An easy default. Room to choose.

Dashy is evaluating OpenAI, Anthropic and Google services. Exact launch models, supported tools, credit costs and subscription plans are not finalized. An external Claude or ChatGPT connection is different from choosing a cloud model inside Dashy Agent.

On-device AI: Supported local features run on compatible devices; browser and model availability apply.
Cloud AI: Choose from the models available in your Dashy release. Cloud requests have account and usage requirements.
Connected assistants: Claude and ChatGPT use their own model and account settings when connected to Dashy tools through MCP.

Capabilities vary by model and configuration. The new picker can show compatible alternatives after a request is rejected. Choosing one starts a new chat with default thinking and preserves the previous chat; it does not send or retry the request automatically. Do not assume every model supports voice, images, search, files, and tools together.

Your controls

Helpful when you want it. Off when you do not.

AI is optional

The new build enables AI assistance by default. Settings → AI assistance turns processing off and cancels active AI work. Your saved tasks, notes, pages, and widgets remain available for manual use.

Pause automatic help

Background assistance can be paused independently. Re-enabling it does not automatically resend a stopped chat or replay a dashboard action.

Review consequential changes

Review setup plans, generated widgets, and proposed edits before applying them. Stop cancels ongoing processing, but it cannot reverse an action that already finished. Undo is available only for supported changes and can be blocked by newer edits.

Choose page and tool access

Page context is removable before Send, protected-site rules apply, and tools have their own permissions. The planned optional page launcher is not an automatic popup. Ask Dashy is also available through supported shortcut, selection-menu, and side-panel entry points in the development build.

Before you use it

What depends on your setup.

The new Dashy build is required for the workflows described here. Installing the original store version does not unlock the complete new AI package.
Cloud Agent features require an eligible signed-in account, a configured service, and network access. Local dashboard essentials work without cloud AI.
On-device AI requires a supported browser, device, language, and available model. Some explicit native-AI actions may require a download. Passive background helpers skip unavailable models.
PDF support currently means locally extracted text with source/page information. It does not imply OCR, scanned-document reading, or native visual PDF interpretation. Images have separate limits and route requirements.
Local account and sync checks do not establish provider quality, production deployment, or compatibility with every device and older extension version. Release availability must be checked in the installed app.

Read the plain-text product guide

Questions people ask

Is the complete Dashy AI package available now?: No. This page describes the new Dashy in development. Several local workflows have been tested, but additional providers, voice, and the full release still have verification gates. The original store version is a separate release.
Do I need to use AI to use Dashy?: No. AI assistance can be disabled. Saved dashboards, tasks, notes, pages, and widgets remain usable manually. Resume on the dashboard can also read and open saved items without a model call.
Does all my data stay on my device?: On-device helpers process supported requests locally. Cloud AI sends the context needed for the action to the configured service, and connected tools may send data to their own services. Review context, permissions, and the privacy policy before using a cloud feature. Local image bytes are not automatically available on another device.
How do I choose a different AI model?: Use the model picker available in your installed Dashy release. Choices depend on account access and the capabilities required for the task. For an external Claude or ChatGPT connection, choose the model in that assistant instead. The final Dashy launch catalog and usage prices will be published before release.
Can every AI model use every Dashy feature?: No. Models and services differ. Dashy checks the required capabilities and can offer a compatible alternative. Changing providers does not make missing credentials, unsupported inputs, or an unverified feature combination available.
Can I undo anything the Agent does?: No. Supported dashboard changes have review and conditional undo, but external actions and newer edits may not be reversible. Stop prevents further processing; it is not a universal undo button.
Do saved chats, memories, and comparison notes sync?: Local tests of the new build verify saved-chat and memory sync across two signed-in Chrome installations, including offline deletions that stay deleted when an older local copy reconnects. Editable comparison notes and cell changes also sync in both directions. These checks use an account eligible for cloud sync. Excluded project memories stay local while that exclusion is enabled, and image attachments may need to be attached again on another device. This is development-build verification, not a promise of compatibility with the original extension or every device.
Will background assistance spend credits by itself?: The current automatic chat-title, recap, and capture-suggestion paths use available on-device AI without a paid fallback. Reusable routines run only after you open and send their instructions. Final plans and credit allowances have not been announced.

Follow the new Dashy

A dashboard that helps you move forward.

Get updates as the new Dashy and its AI features become ready. Tell us what would be useful to you.

Try the original Dashy today4.710,000+ users