Changelog

What if AI could make proving an idea as cheap as it made writing code?

Tools for coding showed that AI could multiply an engineer’s output so greatly, that producing code has stopped being a bottleneck. Grit brings that same leverage to research. It is a single workspace where your sources, PDFs, data, notes, links, and video, stay in one place and become a part of every conversation.

You chat with those sources, and the agent turns what it finds into real outputs: notes, reproducible figures, and reports, written back into your notebooks, so your research compounds enabling you to run experiments, and prove ideas at the speed of AI.

Chat is all you need

Grit centers your work around chat, making it the main view on the desktop.

Chats can live at the workspace level when you want all or no context, or you can place them inside a notebook when a conversation should be grounded in specific sources. You choose where each chat lives, and you can run multiple threads when you need to push more than one idea forward at once.

Modes

Work with Grit in Ask or Agent mode.

Ask is read-only. In ask mode, Grit can search for and read your notebooks content, look things up on the web, and watch YouTube links you include, then answer based on what it finds. It never moves files, writes pages, or changes a notebook, so you get grounded answers without side effects.

Agent can take actions. In Agent mode, Grit can create and edit pages and add sources to notebooks, so you can start research from a question even if you do not have the data yet. It still uses web search and YouTube when needed, combining what it finds with what is in your workspace.

Use Ask when you just want an answer, and Agent when you want Grit to do the work with you.

Switch modes with ⌘Q and ⌘A.

Models

Grit is model agnostic, letting you pick the model you believe fits the work.

The model selector sits at the top of chat so you can switch between fast, inexpensive runs and slower, higher-capacity models when a question needs more depth. You can change models mid-thread as your research moves from quick scans to detailed analysis, and Grit keeps the same context so you do not have to start over.

Ground agents in your sources

The best way to research with AI is to ground it in the information that matters to you. In 0.1.0, Grit lets you add your own sources to notebooks, you can also ask Grit to add sources for you, whether to fill gaps in your information or if you are starting from a question. That way agents can work from your PDFs, data, and notes without you reattaching them in every message.

Upload to notebooks

Grit uses notebooks to organize your work. Add PDFs, notes, spreadsheets, and links as sources in a notebook by dragging and dropping them or using the context menu. Once they’re in, agents can keep using those sources across threads without you reattaching the same files.

Upload to chat

Upload files or links through chat when an item is only relevant for the next message, like a PDF you want quick analysis on or a spreadsheet you want checked. The agent can reference it in that thread, if you ever want it to last, the agent can move sources from any chat into a notebook.

Supported sources

The same types work in a notebook or on a message: PDFs, images, audio, CSV and Excel, web links, and YouTube.

@Mentions

Notebooks and chat already give the agent a lot of context. When something is not in the thread (a page, another chat, a specific file), type @ and mention it.

The mention is inserted inline and included the same way as context from the menu. That is how you point the agent at one source without dumping the whole notebook into the prompt.

Generate artifacts

What good is an agent that cannot close the loop?

Artifact generation means Grit's agent can produce the output that proves the research you are doing matters (a figure, a table, a PDF, a deck) as a durable file in the same notebook as the sources, not a one-off blob in the chat.

Because the agent can generate artifacts, it can also do the ETL that gets you there. Merge CSV and Excel exports, compute columns, flag rows, and write a derived table back into the workspace.

Capture your thoughts

Grit’s agent can create and edit pages for you as your research progresses, keeping reviews and notes organized so they feed back into the research itself.

If you want to capture notes manually, you can use slash commands to insert headings, lists, quotes, tables, images, files, code, and other blocks as you write. Pages in Grit are full documents, not scratchpads, so your notes and sources can stay in one place instead of scattered across apps.

Pay as you go

Grit uses a pay-as-you-go model so you only pay for the inference you use. You can sign up without a credit card, add credits when you need them, and see your spend in the dashboard at gritai.app. There are no seats or long-term contracts to negotiate before you start working, and the credits you purchase never expire.