Skip to content
Guth
Labs · with Tilly, the research and experiment agent

Find out what actually works.

Labs is where you test AI before you trust it. Tilly, the research and experiment agent in your fleet, helps you turn a question into a fair test, compare models or prompts side by side, and read the results in plain words.

Labs app: private preview Instruments and field guides: live

Use these today, free

No account needed.

Live

Instruments
i

A small, documented check or tool, like checking colour contrast or a file checksum. Each page says what it does and how far it has been tested.

A public catalog of small, documented checks that people and agents can run.

Live

Field guides
i

A reusable how-to for one topic, such as coding, security or Windows. Agents load the guides that fit the job.

Practical how-to guides for coding, interface design, security, Windows, macOS and integrations.

Live

For your AI app

Connect any compatible AI app to our public, read-only tools server and use the same guides and instruments.

Research with Tilly
i

A repeatable test that runs different models, prompts or settings on the same examples, scores them the same way, and records the results.

In the private preview, the Labs app is limited and mostly read-only. This is what it is being built to do.

  1. Ask a question“Which model writes better summaries of my notes?”
  2. Set up a fair testTilly picks the comparison, the examples and how to score them.
  3. Run itEvery option gets the same examples and settings, so the result is repeatable.
  4. Read the resultsPlain-language findings, with the failures and the evidence.
  5. Keep a journalEvery test is saved to the shared memory, so you and your other agents can come back and build on it.

Personal model training, made simple In development

Teach a model a specific job, or one of your agents, from your own examples. Guth handles the steps, so you do not need to be an engineer. The trained model joins the same memory system as the rest of your fleet.

  1. CollectGather good examples and sort them by purpose.
  2. CheckConfirm quality, consent and where each example came from.
  3. Train and testTrain, then test on examples it has never seen.
  4. Switch on only if betterCompare with the old behaviour and use it only when it really helped.

Also in development: training and running local models across your own PCs, Macs and even iPhones, working together. Every device, one fleet.

Work with Tilly early

The Labs app is part of the Guth private preview. The instruments and field guides are open to everyone now. Short Agent University lessons are coming soon.