Hugging Face's logo

Spaces:
onitsche
/
talk-with-perplexity
Running

App Files Community
talk-with-perplexity / plan.md
Oliver Nitsche
Initial commit: talk_with_perplexity Reachy Mini JS app
aae9801
preview code
|
raw
history blame contribute delete
3.15 kB

Plan: talk_with_perplexity

What I understand you want

A Hugging Face Space (static JS app) where a user can have a spoken conversation with their Reachy Mini robot, powered by the Perplexity AI API. The robot will respond expressively — moving its head and antennas — while listening, thinking, and speaking.

App Flow 

1. Open Space URL
2. Sign in with HuggingFace (OAuth)
3. On first use: enter your Perplexity API key → saved to localStorage
4. Connect to signaling server → select robot → start session
5. Press TALK button
   └─ Listening  → Web SpeechRecognition transcribes voice → robot listens (antennas up)
   └─ Thinking   → text sent to Perplexity API → robot "thinks" (antennas wiggle)
   └─ Speaking   → Perplexity response read via SpeechSynthesis → robot nods head
6. Conversation history panel shows exchanges
7. Repeat

Technical Approach

Layer Implementation
Hosting Static HF Space (no backend)
Auth HF OAuth (built into the Space)
Robot control reachy-mini.js v1.7.1 via jsDelivr
STT Web Speech API (SpeechRecognition) — browser-native, free
LLM Perplexity /chat/completions — llama-3.1-sonar-small-128k-online model
TTS Web Speech API (speechSynthesis) — browser-native, free
API key User-entered in Settings drawer, stored in localStorage
Robot motion Head/antenna animations during listen/think/speak states

Files: README.md (Space YAML), index.html, style.css

Clarifying Questions

Answer these before I code:

Q1: Talk button style?

  • Push-to-hold (PTT) — press and hold to speak, release to send
  • Click-to-toggle — click once to start listening, click again to stop
  • Auto voice-activity detection (browser auto-stops when you stop speaking)

Q2: Robot camera visible?

  • Yes — show the robot's camera feed (16:9 panel at top)
  • No — conversation-only UI (simpler, faster)

Q3: Robot persona / system prompt?

  • Generic assistant (no special persona)
  • Specific persona: _______

Q4: Perplexity search depth?

  • llama-3.1-sonar-small-128k-online — fast, search-enhanced (default)
  • llama-3.1-sonar-large-128k-online — slower, more thorough
  • llama-3.1-sonar-huge-128k-online — most powerful

Q5: Robot expressive motion?

  • Yes — head/antenna animations for listening / thinking / speaking states
  • No — keep robot still (pure conversation, no motion)

Default Answers (I'll use these if you don't reply)

  • Q1: Auto voice-activity detection
  • Q2: Yes, show camera
  • Q3: Generic assistant
  • Q4: sonar-small (fast)
  • Q5: Yes, expressive motion

Deployment Steps (for later)

  1. hf auth login (one-time)
  2. hf repos create talk-with-perplexity --repo-type space --space-sdk static
  3. git clone https://huggingface.co/spaces/<username>/talk-with-perplexity
  4. Copy files → commit → push
  5. App live at https://<username>-talk-with-perplexity.static.hf.space/