Help

Per period: Free — 500K tokens, deep research 1×. Researcher at €20/month — 8M tokens, deep research 5×. Pro at €50/month — 3M tokens, deep research 20×. Lab at €90/month — 10M tokens plus a shared seat pool for a research group. Settings → Account shows your usage against your allowance.

Two options. Buy a credit pack from the Credits area of Settings → Account: they never expire and are used before your monthly quota. Or turn on overage in Settings → Account, which lets you keep working and bills the excess at your plan's rate. Overage is off by default so you cannot be surprised by a bill.

On desktop, two ways: the built-in terminal for shell commands, and Jupyter notebooks for Python, R and Julia. I can start a kernel, run cells, read the output including errors, and fix the code with you. Terminal commands ask your permission each time. Neither is available in the web app.

Yours. Terminal commands and notebook kernels run on your own machine, or on the server you connected to over SSH. Kosmico does not execute your code on our infrastructure.

Yes — create an SSH remote project. Kosmico reads and writes files on the remote host and runs your terminal commands in a remote shell. Desktop only.

Open the project and click Share in the top toolbar. Invite by email — Kosmico finds existing users as you type — or create a share link, choosing the role either grants. Your co-author gets an email and an in-app notification, and joins as soon as they accept. Cloud sync has to be on for the project.

It depends on the role you gave them. An Owner can do everything, including inviting others and deleting the project. An Editor can edit existing files and create new ones. A Viewer has read-only access. A project can have several owners, and the original creator cannot be removed.

Per project: 1 on Free, 5 on Researcher, 10 on Pro, and 20 on Lab.

There is a full Git panel in the sidebar — stage files, write commits, browse history, create and switch branches, stash, and fetch/pull/push to your remote. If the folder is not a repository yet, the panel offers to initialise one.

Git and Kosmico sharing are separate systems, which is worth knowing before it surprises you. Collaborators see the working tree as it currently stands in the synced project, not your branch structure — so switching branches changes what they see. Their live edits land in the working tree like your own unsaved work; someone still has to commit them.

Kosmico ships an MCP server, so another AI tool can work inside your project. For Claude Code: claude mcp add kosmico -- npx -y @kosmico/mcp-server. For Cursor and VS Code, drop the provided JSON into .cursor/mcp.json or .vscode/mcp.json — Settings → AI has the exact config for each with a copy button. The other tool then gets Kosmico's tools: your files, bibliography, research and notebooks.

Three conditions, and the answer is no on the web app. It is desktop only — the server runs inside the desktop app and there is nothing to start in a browser. Kosmico has to be running, with the project open, because the other tool connects to a local server this app hosts. And it reaches a project that has a folder on your machine — one that lives only in our cloud has no directory for the tools to work in, so it never appears in the list.

I work inside your project rather than talking about it. I read, search and edit your files; search academic databases and manage your bibliography; write and compile LaTeX; edit Word documents and build slide decks; run terminal commands and Jupyter cells on desktop; search the web; and keep your dashboard and notes current. In Agent mode I make changes as I go; in Plan mode I show you the steps first; in Chat mode I only talk. Ask me for the thing — I will tell you if I can't.

Chat answers questions and changes nothing. Agent carries out the work as it goes — editing files, running tools. Plan shows you the steps first and waits for your go-ahead. Switch with the controls under the chat box; use Plan when you want a review step before anything changes.

Yes, through project memory — running notes about the project that I read at the start of every conversation. You can read and edit them yourself at Settings → AI → Project Memory, as a form or as raw Markdown.

Yes — behaviors are custom instruction sets you write. Create them in Settings → AI and pick the active one from the controls under the chat box. Useful for "cite everything and stay formal" versus "be terse and skip the preamble".

From the team roster. Each member has a token allocation you set, drawn from the lab's shared pool and added on top of whatever their own plan gives them — so a student on a Free plan can still work on your budget. The roster also shows seats used against your limit of 20, and you invite by email choosing editor or viewer, which also sets how many synced projects they may create.

The team roster shows each member's project count, their AI usage this period, and when they were last active. That is activity and usage — not their file contents. To read someone's work they have to share the project with you, like any other collaborator.

Not to power the chat here — a ChatGPT Plus or Claude Pro subscription includes no API access, so no other app can route requests through it. Kosmico includes model access in your plan instead: pick any cloud model from the selector in the chat input, or run one locally with Ollama on the desktop app. There is no provider to configure and no key to paste.

The other direction does work, and it is probably what you want: Kosmico ships an MCP server, so Claude Code, Claude Desktop, Cursor or Windsurf — running on the subscription you already pay for — can open this project and use your files, bibliography, research and notebooks as tools. Turn it on in Settings → AI, which has the config for each. Desktop only, with Kosmico running and the project open from a folder on your machine — the other tool talks to a local server this app hosts, so there is no web equivalent.

No. Kosmico serves cloud models through its own backend and there is nowhere to enter an API key. Usage is metered as tokens against your plan. To spend less, pick a smaller model — or a local one, which is free.

Desktop only. Install Ollama and leave it running at localhost:11434 — Kosmico detects it at startup. Then open Settings → AI → Local AI, pick a model (Kosmico checks your GPU, CPU and RAM and highlights the ones that will run well), and download it. Local models cost nothing, work with no internet, and your data never leaves the machine.

Settings → AI → Model selection is where you choose the model. It applies to chat, the floating assistant and the chat widget, and you can set how hard the model thinks there too. If you would rather switch mid-conversation, turn on Show in chat on the same screen and a model picker appears in the chat input. Cloud models are served by Kosmico and included in your plan; on desktop there is also a Local section for models you have downloaded through Ollama. Free plans get the free cloud models; paid plans unlock the rest.

Here is what I can tell you for certain. In a local desktop project your files stay on your own disk; turning on cloud sync or sharing copies their contents to our cloud so collaborators and the web app can reach them. When you chat with me, your message and whatever I read to answer it go to the model you selected — with a local Ollama model, nothing leaves your machine. For anything beyond that — retention, who can access what, whether data is used for training — read the privacy policy at kosmico.ai/privacy or ask support, rather than taking my word for it.

I am not the right source for that answer, and I would rather say so than guess: it depends on company policy and on the terms of the model providers we route to. Check kosmico.ai/privacy or ask support. What I can say is that a local model on the desktop app sends nothing anywhere at all.

Yes, and account deletion is irreversible. Settings → Account → Delete my account asks you to type a confirmation phrase, then permanently removes your profile, sessions, sign-in connections, community activity, and every project you own along with its documents and editing history. Files on your own disk are untouched — delete those yourself. Settings → Account → Export my data downloads your personal data as JSON first.

What the lab roster shows a PI is usage and activity — project counts, AI usage this period, last active — not the contents of your files or your conversations with me. A PI who is also a collaborator on a project sees that project the way any collaborator does. For the formal policy rather than my description of the screen, ask support.

There is no clone button. Clone the repository yourself with git clone — from your own terminal, or from Kosmico's built-in terminal on desktop — then create a project pointing at that folder. Once it is open, the Git panel handles branches, commits and push/pull normally.

Point Kosmico at the folder. When you create a project, choose an existing directory instead of a new one — your files stay exactly where they are, and nothing is copied or converted.

Zotero, yes — on desktop, Kosmico reads your local Zotero library and lets you pick collections to import into your bibliography. Mendeley has no direct integration: export a .bib from it and import that, which Kosmico reads.

The limit counts web projects and cloud-synced ones: 1 on Free, 5 on Researcher, 10 on Pro, 20 on Lab. Purely local desktop projects that you never sync do not count — you can keep as many of those as you like.

Six, in parallel: arXiv for preprints, Semantic Scholar for citation graphs and influence, PubMed for biomedical work, OpenAlex for open metadata across disciplines, CrossRef for the DOI registry, and DBLP for computer science. Results are deduplicated and merged before filtering. Ordinary paper search uses the first four.

Verify them. Every citation I add comes from a record fetched from a real database, and you can open it. But language models do fabricate plausible references, and a wrong citation in your paper is your problem rather than mine — so check anything you are about to publish.

Yes — that is what deep research is for. It builds a plan, searches all six databases, filters and clusters the results by theme, and writes a synthesis, with a live panel you can follow and steer. Per period: Free 1×, Researcher 5×, Pro 20×.

The bibliography panel holds your references. Add them from search results, fetch by DOI or arXiv ID, or import a .bib file. On desktop you can also pull collections from your local Zotero library. Cite from it while you write, and in LaTeX the keys resolve against your .bib.

Seeing structure you would otherwise miss — which papers cluster together, which authors recur, which of your files connect to what you have been reading. Open it from the sidebar or the bibliography panel's graph icon, in 2D or 3D.

Yes — cloud sync is a switch on each project's card in the launcher. Leave it off and the project's contents never leave your machine. You can turn it off again later on projects you own, but that cuts off anyone you shared it with. The switch is unavailable on projects shared *with* you, on team projects your PI has locked, and once you have hit your plan's synced-project limit. Sync is all-or-nothing per project — you cannot choose individual files.

It is a grid of widgets you arrange yourself — tasks, notes, calendar, timers, countdowns, quick links, diagrams, and an AI chat widget. Enter Edit mode to add, remove and rearrange them. I can update most of them for you: ask me to add a todo or a reminder and it appears.

The search bar searches file contents across the project, not just filenames. I can also search semantically when you ask me something — I find the relevant files rather than needing you to name them.

Desktop does everything; the web app lacks the parts that need your machine. Desktop only: local files, SSH remote projects, the terminal, Jupyter notebooks, local Ollama models, the MCP server other AI tools connect to, the local research library, and importing from a local Zotero library. Everything else — writing, collaboration, research, the dashboard — works in both.

Create a project with the Paper template — it sets up the LaTeX editor with a live PDF preview and the bibliography panel. Write, compile in place, and the PDF updates beside you. Compile errors surface inline, and I can fix them with you.

Yes. Kosmico edits .docx directly: I can create documents, read them, navigate by outline, apply formatting and paragraph styles, insert comments, and leave suggested edits rather than silent changes.

Select text and add a review comment. Your co-author can reply in the thread, and you resolve it when it is handled. Mention someone to notify them. I can also review a document and leave comments with concrete suggested edits you accept or reject.

Yes. I can create a deck, write and edit slides, reorder them, set a theme, add speaker notes and images, render a slide to check how it looks, and critique the deck as a whole.