A tour of Studio
The pages of Fernfly Studio and the tabs of a project, and what each one is for.
Fernfly Studio is the signed-in part of Fernfly, where you build, deploy and improve models. This page names each part of it, so the other guides can point to them.
The sidebar
| Item | What it shows |
|---|---|
| Dashboard | Totals across your projects (projects, training pairs, live endpoints and requests served), Estimated savings (what your requests would have cost on a hosted model billed per request), and your top projects. |
| Projects | Every project you own, with its status. |
| New project | Starts the new-project wizard. See Create a model. |
| Profile | Your plan, today's usage against your daily limits, and your sign-in settings. See Usage limits and Account and billing. |
Project statuses
A project's status shows how far it has got.
| Status | Meaning |
|---|---|
| Draft | Created, with no tools yet. |
| Extracted | Has tools, and no training pairs yet. |
| Generating | Writing training pairs. |
| Generated | Has training pairs, and no trained model yet. |
| Training | A training run is in progress. |
| Trained | A run finished, and its model isn't deployed yet. |
| Deployed | A model is live at the project's endpoint. |
A project left at Draft or Extracted shows Finish setting up this project on its Overview, with a Continue setup button that reopens the wizard where you left off.
A project's tabs
| Tab | What it is for |
|---|---|
| Overview | The project's endpoint, its API key, a Try it box, automated tests, the intent bar setup, and the AI fallback setting. |
| Data | The project's tools and training pairs. Export the pairs as JSONL from here. See Training data. |
| History | Every request the deployed model answered: the time, the utterance, the calls it returned and how long it took. Rows used for training are marked in training set. |
| Analytics | Request volume, response times, unanswered requests and the most used tools, over a period you choose. A Pro feature. See Analytics. |
| Retrain | Where you turn real requests into new training pairs and start another run. A Pro feature. See Retrain. |
| Training jobs | Every training run, its log and loss curve, and the controls to stop a run, keep or discard a partial model, and deploy a finished one. See Train and deploy. |
On a Free plan, Analytics and Retrain show a lock. Opening either explains what the feature does and links to the plans.
The Overview, in more detail
When a model is deployed, the Overview shows:
- Active endpoint, with two URLs: Stable (survives redeployments), which always reaches the newest deployment, and Versioned (this deployment only). See Call your model.
- Show code, with ready-made requests in Node.js, Python, Go and curl.
- Show schema, the request and response format.
- API key, the versioned endpoint's key, with Regenerate. See API keys and keyless calls.
- Try it, which sends one utterance to the model and shows the calls and the response time.
- Automated tests, which run up to 20 of your training pairs against the live endpoint and report how many got the right tool, and how many of those also got the right arguments.
- Add the intent bar, the three steps that put the intent bar in your app. See Build an intent bar.
- Settings, with the AI fallback switch. See Call your model.
Requests from Try it and Automated tests are real requests. They appear in History and count toward your daily limits.
Revised