The Conversations page shows the complete history of every conversation users have had with your AI: full transcripts, connected wallet addresses, chains, timestamps, and feedback.
What's recorded
Full transcript
Every message the user sent and every AI response, in order.
Wallet address
The connected wallet address, if the user connected one during the session.
Chain
Which chain was active when the user connected their wallet.
Timestamp
When the conversation started.
Feedback
Whether the user gave a thumbs up or thumbs down on the last AI response.
Session ID
A unique identifier for the conversation, used to correlate with tickets.
The record behind each answer
Every AI answer carries the conditions it was produced under, opened from the shield beside it. This is the part a compliance reviewer reads: the transcript says what was said, the record says whether it can be reproduced.
Chain state
The ledger version the answer was true as of, so the exact chain state can be replayed later.
Prices at read time
The prices any figure rested on. "You were down $312" cannot be checked later without them.
What was checked
The lookups the investigation ran, and any that failed, so a thin answer is never mistaken for a complete one.
Request context
Country, coarse device, surface and language. No IP address is stored in the record. Infrastructure providers process IPs to route and rate limit requests, under their own retention policies.
Model and latency
Which model answered and how long it took.
Answer hash
A SHA-256 of the answer, so any later change to the stored text is detectable.
Records cannot be rewritten
Message content and evidence are append-only, enforced in the database rather than the application. Deletion is refused unless erasure is explicitly requested, and an erasure leaves a tombstone recording who did it and when, so a gap in the record is never unexplained. Views, exports and erasures are logged.
Using conversation history
Conversation history is one of the highest-signal inputs for improving your AI's performance:
- Find questions the AI struggled with: add better documentation to your Knowledge Base
- Spot confident-but-incorrect answers: update your docs or add missing contract context
- Identify recurring problems that might be product issues rather than support issues
- See which chains your users are most active on to prioritise chain configuration
- Understand the language your users use to describe their problems (useful for FAQ block wording)
Review weekly
Spending 10 minutes reading recent conversations each week surfaces knowledge gaps faster than any dashboard metric. Look for responses with many caveats, or questions the user had to rephrase several times.