—
Mentors on-chain
—
Total queries
7
Move modules
2
Walrus data layers
The problem
Agents should not forget everything when the session ends.
Private files, user context, and memory need a durable data layer instead of another app-specific database.
Plain chatbots
Stateless by default. Every important conversation starts from zero.
App-specific memory
Locked to one tool, model, or backend database.
Centralized file storage
Private knowledge can be revoked, hidden, or silently changed.
LLM fine-tunes
Operator can exfiltrate. No clear access policy or memory portability.
Walrus stores encrypted knowledge · MemWal stores agent memory · Sui + Seal enforce verifiable access.
Memory that survives the tab closing.
Knowledge that stays encrypted on Walrus.
Access enforced by Sui and Seal.
The actors
Persistent agents need
persistent data.
Give private expertise durable memory.
Mint a Sui-native AI mentor, upload encrypted knowledge to Walrus, and let every user-specific conversation persist through MemWal. The mentor becomes an agent that can improve over time instead of resetting every session.
Co-own access to a living mentor.
Access shares trade on a bonding curve and gate queries through Sui + Seal. Query revenue is split on-chain, while public confidence and gap signals show whether the mentor is getting sharper or going stale.
Ask an agent that remembers you.
One share gates your access. The mentor answers from encrypted Walrus knowledge plus your own prior conversations from MemWal, so context survives across sessions without being locked into a single app.
The flow
From wallet to
remembered answer.
Mint Mentor
Expert mints a Sui MentorNFT paired with shared MentorState. The object anchors identity, access pools, revenue, confidence, and the current Walrus blob pointer.
Upload Knowledge
Private expertise is encrypted with Seal and written to Walrus. Only the blob id is stored on-chain; decryption is approved by a Move policy.
Buy Access Share
A user buys access through the bonding curve. Holding a share lets Seal key-servers approve decryption under the mentor's policy.
Recall Memory
Before inference, the backend recalls relevant past exchanges for this exact mentor and wallet from MemWal, scoped by namespace.
Answer + Improve
The model answers from Walrus knowledge plus MemWal memory. Low confidence increments an on-chain gap signal; mentor patches rotate the Walrus blob.
Walrus stack integration
Structurally inseparable
from Sui, Walrus, and Seal.
7 Move modules
published on testnet
Mentor identity, per-mentor bonding-curve markets, atomic revenue split, vesting + clawback, and the Seal access policy all live as capability-gated Move objects — no address-whitelist patterns, no privileged backend state.
Published · June 2026Encrypted knowledge vault
Mentor knowledge as a Seal-encrypted blob; only the blob id is anchored on-chain.
Policy-gated decryption
seal_approve checks share balance, oracle, or allow-list — enforced by Seal's key-servers, not Tacit.
Cross-session agent memory
Every exchange is written per (mentor, querier) and recalled before the next answer.
Confidential TEE inference
Architected for Atoma's attested network; swappable to an OpenRouter dev fallback — never reports a fake attestation.
INVARIANT
“No party — not even Tacit's own backend — ever holds a master decryption key.”
Sui Overflow 2026 · Walrus Track
Your knowledge
is an asset.
Live on Sui testnet. The first market for knowledge that stays permanently private — and remembers you.