AI module internals

Agent loop, tools, and safety gating.

Core flow

1. Receive message.

2. Load or create a session.

3. Parse intent.

4. Create a plan.

5. Call tools.

6. Validate with safety rules.

7. Return response + tool traces.

Typical tools

• fetch_bridge_quotes

• compare_routes

• get_token_price

• analyze_portfolio

• get_chain_info

Safety rules

• Slippage limits.

• Chain allow/deny lists.

• Suspicious amount checks.

• Engine health gates.