Troubleshooting.

Chalie logs every LLM call and gives you a full usage breakdown in the Brain dashboard.

1. Open the Brain dashboard.
2. Click Cognition in the left sidebar, then select Usage.
3. You'll see four stat cards: Total Tokens, Cache Hit Rate, Today (UTC), and Top Model.

Use the time-window buttons at the top (Hour, Day, Week, Month, Lifetime) to filter the data. The stacked bar chart below breaks usage into Chat tokens (your conversations) and Subconscious tokens (background processing like memory decay and pattern extraction).

If you're using a paid provider like OpenAI or Anthropic, check the Today (UTC) card for daily cost tracking. Switching to a local model via Ollama removes token costs entirely — usage is still tracked but there's no monetary cost.

Voice ships with every install. It isn't an optional add-on — the Kokoro v1.0 voice model comes bundled with the installer and every upgrade, and it runs entirely on your device, so your text is never sent to an external service. There's no Settings toggle to turn it on or off; if voice ever stops working, reinstalling Chalie is the fix.

Speech happens automatically: The moment Chalie finishes a reply, it synthesizes speech for it in the background — before you ask for it. A speaker icon appears on the final reply of each turn; mid-turn progress lines like "let me check…" are never spoken.

Listening to responses: Click the speaker icon to play the already-synthesized recording, with play/pause, ±10s skip, and a progress slider. While synthesis is still running the button fades in and out and can't be pressed; if synthesis fails, the button turns red and stays disabled — that reply has no audio, and pressing it again won't help.

The voice and speed are not currently adjustable through the UI. Chalie's Personality sliders (warmth, mood, expressiveness, curiosity, humour) affect the conversational tone of responses but do not change the TTS voice itself.

The chat window loads 12 messages at a time and caps at 120 messages total. This is by design — it keeps the interface fast. Scroll up near the top of the chat to automatically load older messages in batches of 12.

Your conversations aren't lost. All messages are stored in your local SQLite database. When the context sent to your LLM provider reaches 80% of its token limit, Chalie automatically summarises older turns into a condensed block so the conversation can continue without losing important context.

Finding older conversations:

• Click the Recall button (star icon in the header) to open a search modal. Type a query like "holiday booking" or "conversation about the project" — Chalie searches across your full history to find it.