№ 001 — The signed record of what an AI agent may do
Your AI isabout to actin your name.
Every act leaves a fossil. An Agent Mandate records who authorized it, in what scope, under which law. Readable as prose, executable as code, revocable on record.
The lifecycle of a mandate
intent → proof
Every decision moves through the same path.
- 01Agent
- 02Authorize
- 03Sign · Hash
- 04Base L2 · cNFT
- 05Verify
Built by conversation
Describe it. The agent drafts the agreement.
Mandate scope — drag the cap
Set the authority. Prove what was agreed.
Drag the cap and the mandate rewrites, rehashes and re-signs. This is the record your signer checks before it releases funds, and the record you hold afterwards.
Agent mandate · acme-delegate-bot
2 transactions outside the mandate's scope
Mandate fingerprint · SHA-256
computing…
recomputed live as the cap changes
Example agent activity
An illustration, not a live agent. Ambr never holds your keys, so it cannot stop a transaction. It makes authority provable: a signed, hashed mandate your signer reads before it releases funds, and a permanent record on Base L2 of what was authorised.
Write your first mandate →§ 02 — How it works
Three steps. One API.
Authorize
Spend caps, allowed counterparties, action scopes, time windows. One API call records the boundaries — signed and hash-verifiable, so anyone can check whether an action fell inside them.
Act
When your agent transacts, Ambr issues an Agent Mandate both sides can trust — readable by humans, parsable by agents. No more handshake-by-prompt-injection.
Prove
Every authorized action is recorded on-chain with a cryptographic hash. Prove exactly who decided what, when, and under whose authority.
§ 03 — One agreement, two readers
Human-readable. Agent-parsable.
What your lawyers see
Plain-language agreement
- Format
- Markdown / PDF
- Jurisdiction
- Delaware, US (default)
- Cryptographic tie
- SHA-256 hash
What your agent executes
The same terms, as data
- Format
- Strict JSON
- State
- Base L2 · ERC-721
- Interface
- REST · A2A · MCP
§ 04 — Pricing
Per contract. Credits are cents.
Free tier: 500 credits once, per verified-email key · Starter Pack $49 = 4,900 credits · Scale Pack $199 = 21,900 credits · Team $29/mo = 3,000 credits · Enterprise $4,800/year · live rates at /api/v1/pricing
§ 05 — A2A Protocol
Discoverable by any AI agent.
The discovery surface
A public agent identity.
Ambr exposes a standard A2A Agent Card, so other agents can find it, check its capabilities, and initiate mandate workflows. No human intermediary required.
Ambr
The signed record of what an AI agent may do.
§ 06 — Built for the stack
The agent-commerce stack.
How agents pay. HTTP-native multi-token payments billed per call within the contract terms.
How agents discover Ambr. Agent-to-agent discovery + JSON-RPC, published at /.well-known/agent.json.
Prove who an agent is — privately. Identity attestation via zero-knowledge proofs.
Provisioned on sign-up — the agent self-keys, the principal gets an email-provisioned wallet. Delegate to an agent without MetaMask or a seed phrase.
§ 07 — The record
Your AI will act.
Under whose rules?
Live on Base mainnet. No credit card. Three lines of code to put real limits on your agent, and a verifiable trail behind every decision it makes.