Agent Commerce · A neutral contracting standard
Agents negotiate. People decide.
Your agents transact with counterparties outside your company, but you remain answerable for what they commit to. Subchain lets agents from different companies propose, counter and accept machine-readable terms against one standard — with a human approval gate before any commitment binds, and a record of who authorised what.
The product, not a promise
The gate and the record — as your team will see them.
A commitment above threshold, held until R. Osei decides — the approval gate.
Who proposed, who approved, what was recorded — the audit record.
Market context
Where this layer sits — sized honestly.
The runtime
One standard, from proposal to binding agreement.
AI colleagues operate the runtime within limits humans set. Named humans govern, approve and hold fiduciary duty.
Deterministic negotiation runtime
An offer–counteroffer–acceptance state machine that enforces one lifecycle across counterparties, so failures are reproducible and state does not drift silently.
Saiba maisMachine-readable term templates
A versioned library covering scope, SLAs, liability, data terms and multi-party obligations — a shared vocabulary rather than free-form text.
Saiba maisAuthority limits and approval gates
A designated human governor must approve commitments above defined value, risk or irreversibility thresholds before an agreement binds.
Saiba maisInline policy-as-code checks
Compliance checks run during negotiation, flagging terms that breach configured rules before they reach acceptance.
Saiba maisImmutable audit record
Every proposal, revision, approval and acceptance is written to an immutable log — evidence-quality material for compliance, audit and DIFC dispute resolution.
Saiba maisRuntime and template licensing
The runtime and template library can be licensed to agent-platform providers as an embed, turning adjacent platforms into distribution channels.
Saiba maisThe lifecycle
From first counterparty to defensible record.
Onboard a counterparty once
One integration to the protocol replaces bespoke connections per relationship. From that single integration it can transact with any adopting party.
Agents negotiate within limits
Agents exchange offers against shared templates within pre-set authority limits, with negotiation state tracked and inline compliance applied.
A human clears the gate
Above a value, risk or irreversibility threshold, the runtime halts and routes to the accountable governor. It cannot bind until they explicitly approve.
Retrieve evidence on demand
Compliance teams retrieve the complete, ordered record of a negotiation for audit, regulatory enquiry or DIFC dispute resolution.
What makes it different
The gate sits inside the protocol, not beside it.
Nothing binds without a named sign-off
No high-value, irreversible or policy-ambiguous commitment becomes binding until an accountable human approves it. The runtime surfaces exceptions rather than deciding them.
Configured once, not integrated repeatedly
One connection to the protocol replaces per-counterparty connectors, so maintenance load stops scaling with each new relationship.
A complete record, not a reconstruction
Every step is attributed to an identity in an immutable log — you retrieve evidence rather than rebuild it.
Not a ledger
Subchain is a human-governed agreement protocol and runtime — not a blockchain, a token or a consensus network.
- DIFC-domiciled
- Human approval gates
- Immutable audit record
- Pre-Seed
Plain answers
Asked by counsel and procurement alike.
Can an agent commit us to something before a person approves?
No. Humans set the authority limits and approval routing before any negotiation runs. Any proposed commitment that exceeds a threshold or is policy-ambiguous halts the runtime and escalates to the named accountable governor, who must give explicit approval before the agreement can bind.
Is this a blockchain, a token or a settlement network?
No. Subchain is a human-governed agreement protocol and runtime. It sits above agent identity and settlement primitives and below application-specific agent workflows, as shared contracting fabric.
What record do we get for audit or a dispute?
Every proposal, revision, approval and acceptance is written to an immutable log recording what was agreed and who authorised it — evidence-quality material for internal audit, regulatory enquiry and DIFC dispute resolution.
Is demand for this proven?
We are candid about the open question. Demand is currently inferred from the adjacent integration pain around it, not yet evidenced by signed design partners. Signing the first design partners is the demand proof this stage exists to deliver.
Is automated contract formation even enforceable?
The legal primitive is settled: UNCITRAL's Electronic Communications Convention (Article 12, 2005) recognises contracts formed by automated systems. The wedge is enterprises already deploying agents for external procurement in jurisdictions where automated contract formation is enforceable but agent authority, approval-gating and liability allocation remain unresolved — the residual risks Subchain governs. Authority-to-bind, error and liability allocation, and DIFC-specific treatment of automated contracting remain open legal items requiring dedicated research before they are relied upon.
Design partners · First cohort
Put a gate between your agents and your obligations.
We are signing the first design partners now. Configure a counterparty once; keep authority with named people.