Know an agent is accountable before you deal with it.

Hyfa gives each agent a Proof of Positive Control: a check, built into the products developers ship, that an accountable company is in charge of the agent and what it may do. The company behind it stays private unless it chooses to say who it is.

Request early access Request the whitepaper

Seller’s owner Seller’s agent

A buyer’s agent asks: Do you sell for the company you name? Proven: yes. Owner not revealed.

Your agent is about to pay $4,800 to an agent it cannot identify.

Tessling’s agent $4,800 “Northdrift” agent

Illustrative, with fictional companies: Tessling Clinics’ purchasing agent holds a quote from an agent claiming to sell for Northdrift Compute. A registry lists the seller and a payment rail moves the money; neither shows whether anyone accountable stands behind that agent.

Pay the wrong agent and nobody answers for the $4,800.

Request early access

One integration under the tools your agents already use

Agents find each other in registries, use tools through MCP, coordinate through A2A and pay through protocols such as x402 or AP2. Hyfa sits beneath them, built into your product the way Stripe sits inside a checkout.

You add it once, and companies that have never met can check each other directly.

Your product

Registries, MCP, A2A

Payments: x402, AP2

Hyfa

If one agent’s key leaks, you cut off that branch and keep the root

Northdrift creates its root key and keeps it; Hyfa never sees it. Each node the root derives grows a thread per customer or check. The name comes from the hypha, a fungus’s filament: reputation flows down while privacy flows up, so outsiders cannot map the business.

A leak costs one branch; the company identity never needs rebuilding.

Try both views, or compromise the billing node.

Root key Sales node Billing node Tessling Buyer two SOC 2 audit Payments

Every agent traces back to someone who answers for it

Each agent has at least one controller, and the chain ends at a member, a company or person accountable for it. Agents inherit what their controller proves. We call that a Proof of Positive Control: someone is provably in charge at every step, never assumed to be.

When an order goes wrong, you know exactly whose agent placed it.

Northdrift ComputeMember

Quoting agentController

Sales agent

SOC 2 current
US operator
Verified domain

A record neither side can forge or later disown

If a claim is ever disputed, both signatures are on record, so neither side can deny it.

Northdriftb81f 44d0 e3a9

Auditorb81f 44d0 e3a9

HyfaAnchored

Replay it, or break it:

Claim anchored.

Direct exchange. Northdrift hands its evidence straight to a fictional auditor.
Separate math. Both sides compute the same sealed fingerprint; a mismatch stops it.
Two signatures. Both sign one record.
Two submissions. Each sends Hyfa its own copy.
Hyfa verifies. It never sees the evidence, only that the copies match and the proofs hold.

Check an agent against your policy and never touch its paperwork

A zero-knowledge proof shows a statement is true without revealing the data behind it. Before paying, Tessling’s agent asks Northdrift’s:

Your team gets its answers and stores none of the other company’s documents.

Tessling’s agent Northdrift’s agent Waiting for a question.

Owner: Northdrift Compute, Inc.
Operator location
Hosting region
Signing limit

The owner stays private until it chooses to prove itself.

In development: a spending limit your agent cannot exceed

Not yet available. Tessling will sign a grant naming action, scope, limit and expiry; anything unwritten is forbidden. The agent holds a scoped key, never the money, and either side can revoke the grant, which only the two of them can see. A spending limit is one example of what a grant can bound.

You set in writing what an agent may spend; anything beyond it is refused.

Grant preview

Action: Buy GPU time
Scope: Northdrift Compute
Limit: $5,000 a month
Expires: 30 days after signing

Purchasing agent Northdrift

Try the grant:

Request early access

Proofs from one deal carry to the next, so nobody onboards an agent twice

Verification lives on the agent’s own threads, so each new counterparty checks the same proofs directly, with no shared database.

Each new deal needs less checking, because the agent arrives with more already proven.

Northdrift’s agent

TesslingProvenTaken on faith
Buyer twoProvenTaken on faith
Buyer threeProvenTaken on faith

Four rules that keep every key in its owner’s hands

Keys stay with their owner. Hyfa never generates or holds them.
Nobody is trusted for their role, Hyfa included. Tessling’s agent checked every claim above for itself.
Prove the answer, not the identity. The checks above kept every record sealed.
Invisible to end users. Your users sign in as usual and never see a key.

Agents your developers can ship and your risk team can approve

Developers building agents

Accounts your users truly own; account recovery is in development.
Scoped grants, once they ship.
Verification counterparties check for themselves.

Enterprises deploying agents

Provable ownership, with an audit trail behind every action.
Policy checks answered by proof, with one integration.

Request early access for your agents

We are onboarding a small group of design partners building with autonomous agents. Hyfa runs on public testnets today; there is no mainnet yet. Investors and technical reviewers can use the same form to request the whitepaper and a briefing.