Use the prompt studio

json
{
  "title": "Use the prompt studio",
  "audience": "power_user",
  "summary": "Inspect and tune the prompt assembly before turns. See what the AI sees.",
  "sort_order": 11
}

Use the prompt studio#

The prompt studio shows you the exact prompt the AI receives on each turn — every system block, plugin context, conversation history, and tool definition. You can edit, preview the result, and optionally apply changes as live overrides.

Who can use it#

By default, prompt studio is gated by tenant feature flag (features.prompt_studio.enabled) and a role/group allowlist. Ask your admin if you don't see the button.

Open it#

Chat header → 🧪 Prompt Studio. Right pane swaps to a two-column view:

Inputs (left) — every knob that affects the prompt: system prompt override, sampling parameters, response mode, plugin loadout, target AI.
Preview (right) — the rendered prompt with each block labelled by source and priority.

What you see in the preview#

Every block has a header showing:

Source — system_prompt, plugin:scaiwave.notes, room_summary, prev_session_summary, active_plan, tool_etiquette, etc.
Priority — integer; higher = closer to the user message. Useful when the prompt gets trimmed for budget.
Token count — approximate; lets you see what's dominating the context.

The user message and assistant turns are shown at the bottom, in chronological order, with their own priority bands.

Make a change#

Edit any of the inputs:

System prompt — overrides the AI's default. Use the textarea.
Temperature / top_p / max_tokens — sliders.
Plugins — enable/disable per-plugin.
Trigger text — what to imagine the user typing next, for the preview's user-turn block.
Target AI — pick which AI in a multi-AI room to apply to.

Click Preview — the right column re-renders with your changes applied. Nothing is committed yet.

Apply or run#

Three actions on a previewed override:

Apply to live room — persist as per-(room × AI) overrides. All future turns in this room with this AI use these settings until you clear them.
Run (sandbox) — invoke the AI once with the previewed prompt, see the response inline in the studio. Doesn't affect the live room.
Clear live override — reset to AI defaults for this room.

The applied override panel at the top of the studio shows what's currently live for this room+AI; cleared status is reflected immediately.

What this is good for#

Tuning a system prompt — "I want the AI to default to bulleted lists in this room. What works?"
Debugging a bad response — "Why did it call the wrong tool? Let me see the actual prompt."
Calibrating temperature — "Is this too hot? Run with 0.3 and see."
Plugin loadout experiments — "Does removing the calendar plugin make the AI faster on conversational tasks?"

What it is not#

Not a chat substitute. The sandbox run is for inspection; if you want a real reply that lands in the room, just chat.
Not where you'd build a custom AI from scratch — that's admin-level participant config.
Not a place to put secrets. Anything you type into the system prompt override is stored in Redis under the room's override key; admins and people with the room can see it.

Where to go next#

Concepts: AI engagement — what each knob does.
API: Prompt Studio.