The platform·how it actually works

How it actually works.

Follow one deal from conversation to signature to settlement. AI agents surface agreement. The signed object becomes executable. The chain moves the money. Private proof remains available.

01·Chat·The reasoning layer

Reasoning that converges to a deal.

People and AI agents talk in the same room. A team of reasoning agents — strategic, domain-aware, deliberately critical — joins the conversation and surfaces what you’ve agreed on, what’s still open, what contradicts what.

Every assertion is a claim. Claims move through a lifecycle: proposed, supported, challenged, accepted. The conversation isn’t a transcript — it’s a graph of converging positions.

Two checks keep the room honest: do the terms make structural sense, and have the open challenges been resolved? Both must pass before anything signs.

M
S
2 peers · 2 agents
REASONING
60/40 revenue splitaccepted
2 supports · 0 unresolved
monthly day-1 cadencechallenged
1 support · 0 unresolved
3-month durationaccepted
3 supports · 0 unresolved
Schema convergence66%
ontological ✓epistemic · 1 open
02·Agree·The signing layer

The chat becomes the contract.

When the conversation converges, it doesn’t produce a transcript. It produces a typed object — a DealPlan — with named fields that match the terms you actually discussed: duration, parties, splits, cadences, caps, conditions.

That object hashes to a single canonical fingerprint. Both sides sign the same fingerprint at the same time. The signed agreement is the executable — no translation from chat to PDF to back-office workflow.

The hash is what reaches the chain. If anyone later changes one byte of the DealPlan, the hash doesn’t match. The agreement and the signature are the same object.

Step 1 · the conversation
maria: 3-month partnership, 60/40?
sam: In. Monthly day-1 cap of $250K?
maria: agreed.
Step 2 · the typed object
DealPlan {
type: "Partnership"
duration: "3mo"
split: [60, 40]
cadence: "monthly:1"
cap: 250000
}
Step 3 · the canonical hash, signed
0x9c4f…e21a
SHA-256
mariasigned
samsigned
03·Settle·The chain layer

Money moves where you agreed.

Every complex deal decomposes into a small set of on-chain primitives. Obligations are NFTs (transferable, cancellable, composable). Swaps are atomic delivery-versus-payment. Payments queue along the cadence the chat agreed to.

A single deal can bundle up to ten of these primitives into one on-chain transaction. The bundle is all-or-nothing: either every leg settles or none do. Half-finished deals are impossible by construction.

Need approvals? Declare a signing policy on any wallet — "this kind of operation needs 2 of 3 signers" — and the chain enforces it. The pipeline holds matching operations until the right signatures arrive.

The deal
Acme · 3-mo restructure
Decomposes into primitives
Obligation
NFT #4521
Swap
A↔B atomic
Payments
×3 monthly
ATOMIC · all-or-nothing
Settled on chaintx 0x9c4f…e21a
04·Repeat·The memory layer

Private trust compounds.

Every closed deal leaves a sealed, verifiable receipt shared only with the parties involved. It proves settlement happened without making terms, counterparties, or amounts visible to the world.

When history helps the next deal, the parties choose what to reveal: one receipt, an aggregate repeat-deal proof, or the full agreement. The default is confidentiality; disclosure is deliberate.

This is not a platform reputation score or a public graph. It is portable trust you control — proof that work happened, available only at the level you decide to share.

Private deal memory
37 private receipts
party a
12 receipts
party b
selective proof18 mo
Terms stay private unless both parties choose to disclose them.
party a + party b
private repeat proof
12
receipts
party b + party c
shared privately
7
receipts
party d + party e
new private edge
3
receipts
Receipts stay sealed. Parties choose what proof, if any, to reveal.

Chat. Agree. Settle. Repeat.

Now you’ve seen the mechanism. Apply to the cohort if you’re a dealmaker, or open the docs if you’re building on it.

Cohort 1 · 7 seats · closes 13d
IV of V·End of chapter
YieldFabric docs(317)