AI Response Format

Juno passes structured context into the assistant. The model doesn't see raw txs, logs, or balance sheets, only a scoped prompt based on current wallet state, memory, and filter.

The system message includes:

• Wallet summary

• Signal history

• Token risk flags

• User filter

• Recent prompt history

The assistant is expected to return a short, high-signal summary, no speculation, no repetition.

---

Example Prompt

Wallet 9gzL...vU7W is holding 3 tokens, total value $1,320.

Signals:
- buy-pressure (0.91 confidence, 2h ago)
- trend-reversal (0.78 confidence, 5m ago)

Token BONK is mintable and unverified.

Filter: trend-reversal

Summarize current state and highlight anything noteworthy.

---

Response Expectations

• 1–3 sentences max

• No emojis or filler

• Point to specific behaviors, not general market info

• Include confidence when relevant

---

Assistant doesn’t fetch data. All it does is interpret context passed in.

This keeps latency low and response quality consistent.