What’s new & why it matters

1) Short-circuiting for conversation-enders & inappropriate input

When the user says things like “thanks, that’s all” or when our safety/off-policy checks trigger, AURA now takes a fast exit path:

• Wraps up politely (or blocks with a brief notice if needed)

• Skips unnecessary reasoning/generation

• Reduces latency and wasted tokens

Good for: tighter UX, fewer awkward “one more thing?” replies, and safer behavior.

2) Pronoun resolver for accurate anchoring

AURA resolves pronouns back to the intended entities (speaker, character, referenced person/thing) before composing the reply.

• Fewer identity mix-ups in multi-turn chats

• Better continuity when users switch names/pronouns mid-thread

Example: If we last mentioned John and the user asks “What does he think?”, AURA keeps the context pinned to John unless a new subject is clearly introduced.

3) Clarify-gating for ambiguous framing

If an input could reasonably mean multiple things (or confidence is low), AURA now asks a quick clarifying question first.
Result: fewer wrong turns, shorter paths to the right answer, and nicer transcripts.

4) Prompt hygiene & small performance gains

We removed a handful of legacy prompt fillers that no longer pull their weight.

• Leaner prompt = slightly snappier responses

• Less drift from redundant guidance

5) Deep Dialogue overlay (Dashboard)

Meet the Deep Dialogue overlay: a focused, cinematic chat layer that keeps your logs rolling underneath.

Highlights

• Typewriter subtitles + real-time responses

• Optional TTS playback; Esc to close

• Shows the current drift percentage

• Keyboard friendly; keeps your chat history in sync in the background

Try it

1. Go to Interaction.

2. Click Interact.

3. Chat in the overlay; your turns are logged as usual.

Developer notes

• The overlay respects Canon Strict and Mute Voices toggles.

• Conversation-ender detection is intent-based with safety hooks; ambiguous-input gating uses a lightweight confidence check before asking for clarification.

• No breaking API changes; this is mostly orchestration + UX.

Changelog

• Added short-circuit path for endings & safety.

• Added pronoun resolver to stabilize entity references.

• Added clarify-gating for ambiguous prompts.

• Removed stale prompt fillers (minor speed/clarity win).

• Shipped Deep Dialogue overlay in the dashboard.