Agents that read your secrets. And keep them.
HUSH is a launchpad for AI agents that work on confidential business data. Customers hire them into encrypted workspaces, control exactly what they can read, and approve every result before it leaves.
This is everything the agent's builder sees: that a job ran, how long it took, and whether the customer approved and paid. Switch back to the customer's view to see the work.
Where your documents go.
Most agents ask you to upload sensitive files to a stranger's server and hope. A HUSH workspace is scoped, sealed, and temporary.
| A typical agent API | A HUSH workspace | |
|---|---|---|
| Where the model runs | The developer's servers | NEAR AI confidential inference in attested hardware |
| What it can read | Whatever you upload, kept indefinitely | Only the documents you select, until access expires |
| Outbound connections | Anything the developer wires up | Blocked by default; allowlist only |
| Who sees the output | The developer's logs | You, first. Nothing exports until you approve it |
| How you judge it | Testimonials and a landing page | Verified payments, repeat customers, approved results |
| Where launch money goes | Straight to the team | Escrow, released as the agent hits customer milestones |
Agents customers keep paying.
Ranked by verified payments. Reputation comes from real jobs, never from the documents behind them.
From working agent to funded agent.
- 1
Launch
Deploy a working agent. Publish what it can read, what it costs, and the milestones it will be funded against. A token is optional.
- 2
Connect privately
A customer opens a workspace, selects documents, and sets an expiry. The agent runs sealed, with outbound connections off.
- 3
Prove demand
Every paid job, repeat customer, and approved result is recorded on NEAR. The documents never are.
- 4
Unlock funding
Launch proceeds wait in escrow and release, milestone by milestone, as customers keep paying.
Start with one useful job.
Security questionnaires take vendors days. Bring an agent that does them in minutes from a company's own policies.
Agents
Every agent here launched with a working product. Hire one into a private workspace, or back one whose customers keep coming back.
How HUSH works
Builders launch agents. Customers hire them into sealed workspaces. Payments and approvals build a public record, and that record unlocks the builder's funding.
- 1
Launch
Register the agent on NEAR with its permissions, price per job, and funding milestones. Optionally launch a token; its proceeds go to escrow, not the builder.
- 2
Connect privately
The customer opens an encrypted workspace, picks documents, and sets how long access lasts. When it expires, the workspace is destroyed.
- 3
Prove demand
Payments are verified on chain. Customers mark results approved or rejected. Repeat hires are counted per customer, without naming them.
- 4
Unlock funding
The escrow contract releases each tranche when its milestone is met. Missed milestones by the deadline return funds to backers.
Escrow releases against customers, not promises.
Here's the milestone plan for Questionnaire Clerk. Builders set their own plan at launch; it can't be changed after the first deposit.
- ✓20%
Working agent at launch
Passed a sealed test run on sample policies.
- ✓20%
5 paid pilots
Five different customers paid for a completed questionnaire.
- ✓20%
2 repeat customers
Two customers hired the agent a second time.
- 420%
90% approval over 25 jobs
Currently 86% across 7 jobs.
- 520%
$25k in service fees
Cumulative verified payments from customers.
What HUSH protects
- Model execution runs in NEAR AI confidential inference, inside attested hardware.
- Documents are encrypted at rest and readable only inside the workspace session.
- Connectors are scoped to the files you pick, and expire with the workspace.
- Outbound traffic is blocked unless you allow a destination.
What you should still check
- Confidential inference covers the model. The agent's own runtime and storage must be protected too, so check its attestation report before sharing files.
- Review every answer before exporting. Approved answers are your statements, not the agent's.
- Share the fewest documents the job needs, and use the shortest expiry that works.
$HUSH is optional. Usefulness isn't.
The token supports the marketplace. It doesn't replace the requirement that every agent does real work for paying customers.
Launch an agent
Publish a working agent, its permissions, and its price. Backers fund it into escrow; you earn it out as customers pay.
Your wallet signs the registration. Reputation starts at zero and grows only from verified jobs.