Agent tools & Agent Compute

What agents can do beyond writing — read and search your project, and run code to handle tasks no built-in tool covers.

Agents do more than write. They can read your project, search it, and run code — which is how Multilo turns a request like “graph this CSV” or “pull the tables out of this PDF” into a real result instead of a guess.

Reading & searching

Agents can read the documents in your project — including the text of PDFs and Office files — search across everything you’ve added, and edit your draft. This grounding is what lets them work from your actual sources rather than generic text. See Super Search and files & folders.

Two expanded tool steps under a chat reply: Read Reading notes.md (5 lines) and a search for “burial” that found 5 lines in 2 files. — Every file the agent reads and every search it runs is listed in the chat.

Running code

When no built-in tool fits, an agent can write and run code to get the job done — so almost any file can become almost any outcome: a chart from a CSV, text extracted from a PDF, a converted format, a quick analysis. You don’t set any of this up; the agent decides when code is the right tool and runs it for you.

A card asking permission to run python3 Scripts/plot_burial.py, with a note that the first run sets up Python inside Multilo. — Running a script asks first, unless you have allowed it.

Safe and permissioned

Code runs in a secure, isolated environment and is capped so a long job never freezes the app — even on a very large document. For heavier work, an agent asks your permission before going further, so nothing runs outside what you allow. As always, this happens locally; your files don’t leave your machine except when a model is called.

It just works

You never configure runtimes or environments. Ask for the outcome you want and the agent figures out whether — and how — to use code to get there.