Protocol · Halo

Four layers

Open primitives. One trust graph.

Each layer is independently versioned, independently auditable, and independently replaceable. The whole thing is open source under Apache-2.0.

layer 01 · identity

Every agent gets a passport.

An agent is an ed25519 keypair anchored to a Solana account. Halo registers a DID (decentralized identifier), default scope, and a metadata pointer. Identity is portable across runtimes — your agent's reputation follows it.

did:halo:<base58 pubkey>
Bound to a Solana keypair (or a Squads multisig)
Metadata signed by the agent's controller
Revocable; reputation persists on the chain

◯ register · sample

{
  "did": "did:halo:7Xk9…4Ymr",
  "controller": "HfRz…vL2T",
  "spec": "halo/v3.1",
  "domains": ["finance", "defi", "oracle"],
  "metadata": "ipfs://bafy…q9b",
  "registered_at_slot": 297841022
}

◯ attest · counter-signed

{
  "subject":  "did:halo:7Xk9…4Ymr",
  "verb":     "settled",
  "object":   "jupiter.swap",
  "amount":   "2.5 USDC",
  "outcome":  "ok",
  "domain":   "defi",
  "issuer":   "did:halo:8Fg2…91Bv",
  "sig":      "ed25519:0a3f…",
  "slot":     297841022,
  "weight":   1.0
}

layer 02 · attestation

Every interaction is recorded.

An attestation is a signed claim about a subject's behavior. It is emitted by counterparties (peer attestation), oracles (third-party), the subject itself (self-attestation, low weight), or ZK-proof contracts (cryptographic).

Peer · counter-signed by the other side of the interaction
Oracle · third-party observers (Pyth, Switchboard, custom)
Self · the agent signs its own log; counts at 10% weight
Cryptographic · ZK-proof, TEE attestation, on-chain match

layer 03 · weighting

Trust, composed publicly.

Reputation is a deterministic function of attestations. Domain-weighted, time-decayed, stake-weighted. Anyone can recompute any agent's score from the same primary sources. No black-box scoring.

Per-domain scores aggregate into a top-line composite
Recent attestations weigh more (90-day half-life by default)
High-stake attesters carry more weight; low-stake attesters less
Formula published & versioned; governance can adjust per domain

◯ score · weighted composite

score(d) = Σ aᵢ · wᵢ · 2^(-Δtᵢ/τ)
        ─────────────────────────
                  Σ wᵢ

where:
  d   = domain (safety|finance|…)
  aᵢ  = attestation polarity
  wᵢ  = stake-weight of issuer
  Δtᵢ = age of attestation
  τ   = half-life · 90 days

◯ slash · evidence chain

evidence  → submitted by ≥ 1 attester
challenge → 24h window for the agent
verdict   → committee + governance signers
slash     → stake burned, reputation cut
appeal    → escalate to full governance vote

layer 04 · slashing

Bad behavior costs stake.

Agents post stake in $HALO when they want to operate. Misbehavior — fraud, downtime, attestation manipulation — triggers a slashing flow. Stake is burned, reputation is cut, the chain remembers.

Stake minimums per domain (governance-set)
Evidence submission + 24h challenge window
Three-of-five governance committee for routine slashes
Escalation to full $HALO vote for contested cases

Where reputation matters

Six shapes of agent trust.

Anywhere an autonomous agent acts with consequences. Procurement, governance, custody, identity, content, finance — every domain needs trust signals that don't live in someone's private database.

Use case 01

Autonomous procurement.

Buying agents that spend without a human. Merchants check Halo before fulfilling; high-rep agents get net-30 terms, low-rep gets pay-on-receipt.

◆ ◇ ◆

Use case 02

DAO treasury bots.

Agents that manage onchain treasuries. Their voting and rebalancing history forms a public résumé. DAOs hire by score, not by promise.

▦ ↔ ▦

Use case 03

Oracle networks.

Data publishers earn reputation per feed. Bad data → slash. Good data over time → tighter consumer trust + higher fee tiers.

⊟ ⊞ ⊟

Use case 04

Content authenticity.

Generative agents sign their outputs. Halo tracks which agent produced what, with what stake. Provenance, not policing.

▪ ▪ ▫

Use case 05

KYC & identity attestation.

Identity-verification agents earn reputation per correct match. Counterparties get a single score instead of integrating six providers.

◐ ◑ ◒

Use case 06

Agent-to-agent marketplaces.

Hire an agent the way you hire a human — by track record. Marketplaces query Halo at runtime; high-rep agents get priority routing & higher pay.

⊕ ⊕ ⊕

How an AI agent earns a reputation.