Test drive (debug chat)
The live chat panel inside the character editor. Edit the form, regenerate, see what changed — without ever leaving the page.
The hardest part of building a character used to be the save-and-pray loop: edit form, save, navigate to a new chat, send a message, regret an edit, navigate back, repeat. The debug-chat panel cuts that to one screen.
Source post: Test drive your character →
What it is
A floating chat panel that lives on every character creation surface. It runs your character live, against the current state of every form field. Change a field, hit retry on the last reply, see the difference immediately.
How to use it
Open the panel
In the character editor: floating Debug chat button in the bottom-right. Opens a side panel without closing your form.
Send a message
Type anything — usually your real-world greeting test: "hi", "tell me about yourself", or a deliberately weird prompt that exposes whether the character has range.
Read the reply, then change one field
Edit any field — personality, sample dialogue, system prompt, styling rule — and hit ↺ Retry on the AI's reply. The reply regenerates with the updated field in play.
Iterate
Three-message conversations are usually enough to feel one field's effect. For bigger changes, run a full 10-message arc.
What's different from real chat
| Real chat | Debug chat |
|---|---|
| Costs credits per message | Free for subscribers; small cost for non-subscribers |
| Saves to your conversations | Lives only while you're in the editor |
| Picks the model from your settings | Defaults to a fast model, switchable |
| Counts toward character analytics | Doesn't count |
Limits
The debug panel is unmetered for subscribers. Free-tier users get a daily quota of debug messages — when you hit the cap, the panel tells you and offers an upgrade prompt, or you wait for the next day's reset.
Common iteration loop
- First five messages — gauge whether voice is consistent.
- Edit sample dialogue if voice drifts.
- Push a weird scenario (rude prompt, off-topic question) — gauges range.
- Edit personality refusal list if range is bad.
- Push an emotional scene — gauges depth.
- Edit styling if scene reads thin.
- Send the greeting alone in a fresh chat — gauges first impression.
Repeat until it feels right.
Testing scenarios
The debug panel includes a scenario picker above the input. Switch scenarios mid-test to see how each one performs. This is critical for multi-scenario characters — a scenario that looks fine on the form might fall flat in practice.
Testing different models
The debug panel has a model picker identical to the real chat's. Test on:
- A free or cheap model (Llama 3.1 8B, DeepSeek V4 Flash) — most users start here
- The premium model you'd want users to choose (GLM 5, Gemini 3)
A character that only sings on Gemini will disappoint most users.
Testing with plugins active
The debug panel respects your pre-installed plugins for the character. If you set the character up with a stats panel and dice plugin, those are available in debug-chat. Test the plugin behaviour here before publishing.
When debug-chat doesn't catch a bug
A few classes of bug only show up in real chats:
- Long-term memory — the debug panel session is short. Real chats run for hundreds of messages and reveal memory drift.
- Group chats — the debug panel is solo. Add the character to a real group chat to see how they play with others.
- Story mode — different model, different prompts. Run a story playtest separately.
For those, publish privately, test in real conditions for a day, then switch to public.