Agent Capabilities

It's easy to under-use an agent by not knowing what to ask. This page is the tour of what your agent can actually do out of the box — no setup, no add-ons — so you can calibrate your requests to its real range. The appliance ships with substantially more capability than a stock chat assistant; treat this list as a menu.

Research and the web

The agent can search and read the web: current information, competitor sites, supplier documentation, reviews, industry news. Point it at a question rather than a link — "research current pricing for 24/7 gym access in our area and summarise the spread" — and it will search, read, cross-reference, and write up findings with sources. Pair with the wiki for research you want maintained over time.

Documents, spreadsheets, and PDFs

Give it files — drag them into chat — and it reads them properly:

  • Spreadsheets: analyse, cross-reference, reconcile, pivot. "Compare this month's export with last month's and list members who lapsed."
  • PDFs: contracts, invoices, statements, manuals — it extracts, summarises, and answers questions against the actual text.
  • Word documents and text: review, rewrite, merge, reformat.

It can work across multiple files at once — reconciling an export against an invoice is a one-message request.

Images and OCR

The agent handles images as inputs: photograph a paper form, a whiteboard, a delivery docket, an error on a screen — it reads the text (OCR), interprets what it sees, and acts on it. "Here's a photo of the supplier's price list — put it in a spreadsheet."

Producing files and reports

Outputs aren't limited to chat. Ask for real artifacts and the agent writes them into its workspace and hands you a download: formatted documents, spreadsheets with working formulas, CSV exports, charts and graphs it generates from data, even simple slide-ready summaries. The habit that pays: "…as a file" at the end of a request.

Code and data analysis

Behind the scenes, the agent can write and run code on its own device — which is what makes the heavier requests possible: statistical analysis over a year of attendance data, deduplicating a messy export, converting between file formats, generating a chart. You never see the code unless you ask; you just get the result. (Its autonomy to run things is governed by the command approvals setting.)

Media and speech

The agent can produce spoken audio from text — useful for announcements or accessibility — and process audio and video files you give it, within the practical limits of the device's storage.

Your business, live

The capability that ties the rest together: through its Performance Hub connection, everything above can be done against your live operational data — members, attendance, bookings, devices, payments, reports, across whatever scope you've granted. Analysis of a spreadsheet is useful; the same analysis against this morning's real numbers, unprompted at 7am via an automation, is a different product.

Delegation and long tasks

Big jobs don't need babysitting. The agent can break a large request into parts, work through them — including delegating subtasks internally — and notify you when the result is ready. Ask for the quarterly analysis and go run your facility.

Learning new tricks

Anything it does well can become a repeatable, named procedure — see Skills. The out-of-box capabilities are the raw material; skills are how they compound into your workflows.

Getting the best out of it — prompting that works

Six habits that consistently improve results:

  1. State the goal, not the steps. "I need to know which membership tier is underperforming" beats a list of micro-instructions — let it choose the method.
  2. Give context. Who it's for, what it's for, what good looks like: "for the owners' meeting — one page, numbers first."
  3. Name the output. "As a table", "as a spreadsheet in your workspace", "as three bullet points I can text to staff."
  4. Correct early and specifically. The first draft is a proposal. "Right numbers, wrong tone — make it more formal" gets you there in one round.
  5. Promote what works. The moment a result is exactly right: "save that as a skill."
  6. One topic per conversation. Fresh conversation for fresh work; memory carries the context that matters.