Dashboard
The Dashboard is the builder home at /studio — a dense workspace overview for monitoring agents, models, training, and recent activity.
Top widgets
Section titled “Top widgets”
- Active agents: how many of the project’s agents are running (system agents left out). Its “deploying” count stays at 0, since starting an agent switches it on at once with no deploy phase
- Tokens today — total LLM tokens consumed in the last 24h
- Success Rate — % of sessions completed without crashing or being denied
- Deployed agents: every agent in the project except system agents, with its model and status and a count of the running ones; click an agent to open it. Its replica count usually reads 0 because nothing updates it: agents run on the shared runtime and have no replicas
Lower sections
Section titled “Lower sections”- Serving Models — fine-tuned models currently being served
- Training Jobs — active SFT / GRPO runs
- Eval Runs — active evaluation runs
- Recent Conversations — feed of recent sessions across all agents
- Filter chips: All / Flagged
Quick Actions
Section titled “Quick Actions”Five shortcut buttons:
| Button | What it opens |
|---|---|
| New Dataset | NEW DATASET picker |
| Start Training | The NEW RUN form on the Training page (SFT or GRPO) |
| New Eval | NEW EVALUATION wizard |
| Serve Model | + Deploy model flow (Local Hub / HF / OpenRouter / URL) |
| New Expert | Create an expert (a skill backed by a fine-tuned model) |
Each opens the same modal/wizard you’d reach by navigating to its module page.
Daily ops habit
Section titled “Daily ops habit”A common builder workflow:
- Morning: open the dashboard.
- Glance at success rate and error count. Drill into errors if any.
- Check Recent Conversations → Flagged. Triage the worst sessions.
- Quick Actions when something needs urgent attention (deploy a hot-fix agent, start a small fine-tune).
What’s next
Section titled “What’s next”Copilot for the platform-driving agent. Templates for the deploy flow.