Product brief · NEAR Hackathon 2026 · AI Swarms
An auditable newsroom,not a content farm.
The Daily Swarm is an AI newsroom for a niche beat. Agents pitch stories, an editor commissions two under a budget, and a standards desk checks every claim against a source pack. Readers pay once; the revenue split settles to agent sub-accounts on NEAR.
One edition, four moves
From pitch to proof.
The agents do the work; the colors show who did it. Beat colors mark the three desks, red is the editor, green is the standards desk, and gold is the settlement.
- 01
Pitch
Three desks file a headline, a fee, a word count, and claims tied to the source pack.
- 02
Commission
The editor commissions two under a fixed budget and records why the third was passed.
- 03
Check
The standards desk checks every claim. An unsupported claim forces one revision and a re-check.
- 04
Settle
A reader pays 1 testnet NEAR. The deterministic split settles to agent sub-accounts and lands in the public ledger.
The problem
AI news has a trust problem.
Readers cannot see the work
Which source supports which sentence? What was checked? What was flagged and rewritten? Most AI news products hide all of it.
Agents cannot show the books
Agent economies need a record of work, verification, and payment. A receipt and a public ledger are the missing trust layer.
The product
A daily edition with claim-to-source mapping, independent fact-check reports, a reader receipt, and a deterministic settlement ledger. The edition is the output; the settlement is the proof.
Who it is for
Three audiences, one audit trail.
Crypto-native readers
A daily, sourced brief instead of an opaque feed. Every claim maps to a source, and the edition shows what the standards desk flagged.
Protocols and DAOs
Commissioned coverage with a public audit trail: what was pitched, what was commissioned, what was checked, and what was paid.
Builders and researchers
A working reference implementation of an agent economy: roles, budgets, a hard fact-check gate, and deterministic settlement.
5
agents
48
tests
1
testnet NEAR
In this build
What is real
- ✓Full pipeline: pitch → commission → write → check → revise → re-check → layout.
- ✓Every factual claim must cite a source ID; unsupported claims force a revision.
- ✓Five agents with NEAR testnet sub-accounts and a public ledger.
- ✓48 tests, CI, and a production build that runs without secrets.
- ✓A no-wallet sample edition and a real testnet settlement record.
Still demo
What is next
- —The source pack is curated: six topics, 30 facts. The agents do not browse the web.
- —The public instance does not hold the treasury signing key; payouts are local accounting.
- —Editions live in the browser, so a shared edition link does not travel.
- —Testnet only; no accounts, database, or rate limiting.
The demo is deliberately honest about the seam: curated sources now, a live source-pack builder and a human review console next.
The path
From auditable demo to product.
- 01
Phase 1 · Now
Auditable demo
Prove the full loop and make the audit trail visible. The curated pack keeps the demo stable and every claim verifiable.
- 02
Phase 2 · Next
Live niche newsroom
A source-pack builder for one beat: ingest primary sources, snapshot and hash them, extract claim/span evidence. Add a human review console, a database, and shareable editions.
- 03
Phase 3 · Then
Product
Reader subscriptions or micropayments, a B2B audit/verification API, licensed and sponsored beats, mainnet settlement, and agent reputation.
Business model
The ledger is the trust layer.
The news product is the content; the audit trail is why someone pays for it. Reader revenue, a B2B verification API, and disclosed sponsored beats can share the same pipeline.
Reader revenue
A subscription or micropayment for the daily edition. The public ledger is part of what the reader is paying for.
Audit API
Publishers and agent teams can buy the same source-grounded verification layer: claim → source span → verdict.
Sponsored beats
Protocols can commission a beat with disclosure. The ledger shows exactly what was paid and to which agent.
Moat and risks
Built to be checked.
Source-grounded verification with a hard gate, a deterministic split, and an open reference implementation. The risks are the ones a real newsroom has; the mitigation is the same: sources, review, and corrections.
Copyright and licensing
Use primary sources and licensed feeds. Do not scrape or rewrite paywalled content.
Defamation and privacy
Human review, corrections, and legal review for high-risk beats. Nothing publishes automatically in Phase 2.
Hallucination
Source IDs, span-level evidence, schema validation, and re-checks. An unsupported claim is a visible failure, not a silent rewrite.
Cost and trust
Retrieval, caching, and cheaper models; label AI involvement, show sources, and publish corrections.
See the product
Read the sample edition, then produce a live one.