AI Pipeline
Every AI feature in Spun — Compose, Translate, Summarize, Smart Replies, Chatbot — flows through the same pipeline. Understanding it helps you manage token usage and get better results.
The AI flow
Token economy
Tokens are the currency of AI usage. Every AI action costs tokens:
| Action | Approximate cost | Notes |
|---|---|---|
| Smart Reply (3 suggestions) | ~200 tokens | Cheapest — short context |
| AI Compose (short message) | ~500 tokens | Depends on prompt length |
| AI Translate (message) | ~300 tokens | Per message translated |
| AI Summarize (conversation) | ~1000-3000 tokens | Scales with conversation length |
| Chatbot response | ~800 tokens | Includes knowledge retrieval |
Monthly allowances by plan
| Plan | Monthly tokens | Daily caps |
|---|---|---|
| Free Trial | 500K | None (trial period) |
| Basic | 50K | 2 compose, 10 smart replies, 20 translations, 1 summary |
| Pro | 500K | Unlimited |
| Power | 4M | Unlimited |
When you run out, you can buy token top-up packs. Purchased tokens don't expire.
Context assembly
Before calling the AI model, Spun assembles context:
- Your prompt — what you asked for (compose instruction, translate source, etc.)
- Chat history — recent messages for context (limited to save tokens)
- Contact info — name, labels, custom fields
- Feature knowledge — Weave's knowledge base for chatbot responses
- User preferences — your language, tone settings, AI model preferences
Model selection
Spun uses different models for different tasks:
- Fast tasks (Smart Replies, translation): smaller, faster models
- Complex tasks (Compose, Summarize, Chatbot): larger, more capable models
- Retry with smarter model: if the first response is unsatisfying, click "Retry" for a more capable model
Tips for better AI results
- Be specific in prompts: "Write a friendly 2-sentence follow-up about the proposal" > "Write a reply"
- Provide context: The more relevant chat history, the better the output
- Use the right feature: Smart Replies for quick responses, Compose for thoughtful ones
- Monitor usage: Settings → Billing → Token Usage shows a breakdown by feature
AI Logs (audit trail)
Every AI operation is logged in Settings → AI Logs:
- What was requested
- What was generated
- Tokens consumed
- Timestamp and user
This is your audit trail for all AI activity in the workspace.