Intaglio Documentation

#Quick Start

One HTTP endpoint, no SDK required. Every decision gets a SHA-256 chain link and a stored audit record.

# 1–2. Create the agent, set its limits and mint a key in the dashboard
export INTAGLIO_API_KEY=intg_live_...

# 3. Enforce an action before your agent executes it
curl -X POST https://solanacompliance.com/api/enforce \
  -H "Authorization: Bearer $INTAGLIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent_slug":"my-agent","action":{"type":"transfer","amount":100,"currency":"USDC","destination":"7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJevxJSw"}}'

# 4. Anyone can check a decision by its hash — no key needed
curl https://solanacompliance.com/api/verify/<hash>

#Authentication

POST /api/enforce requires a Bearer token in the Authorization header. Each key belongs to one operator and can only enforce for that operator's agents.

GET /api/verify/{hash} is public and needs no key. Everything else — agents, limits, approvals, audit export — is managed in the dashboard with your signed-in session.

#Enforce an Action

The core primitive. Every agent action passes through this endpoint before execution. Rule evaluation is deterministic and typically takes well under a millisecond (enforcement_ms); end-to-end latency also includes OFAC screening and storage (total_ms).

Request

curl -X POST https://solanacompliance.com/api/enforce \
  -H "Authorization: Bearer $INTAGLIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_slug": "my-defi-agent",
    "action": {
      "type": "transfer",
      "amount": 4200,
      "currency": "USDC",
      "destination": "7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJevxJSw"
    }
  }'

Response — DENY

{
  "outcome": "DENY",
  "reason": "Transaction amount 12000 USDC exceeds daily limit of 10000 USDC",
  "rule_triggered": "hard_block",
  "hash": "sha256:5d1e…",
  "prev_hash": "sha256:9f2c…",
  "enforcement_ms": 0.38,
  "ofac_ms": 21.4,
  "total_ms": 71,
  "audit_record_id": "b0c4…",
  "approval_request_id": null,
  "policy_found": true,
  "risk_score": 70,
  "risk_level": "high",
  "kya_flags": ["LARGE_TRANSACTION"],
  "solana_tx": null,
  "solana_explorer": null
}

Response — APPROVE

{
  "outcome": "APPROVE",
  "reason": "Within policy limits",
  "rule_triggered": "auto_approve",
  "hash": "sha256:9f2c…",
  "prev_hash": "sha256:41ab…",
  "enforcement_ms": 0.41,
  "ofac_ms": 19.8,
  "total_ms": 66,
  "audit_record_id": "3f6c…",
  "approval_request_id": null,
  "policy_found": true,
  "risk_score": 0,
  "risk_level": "low",
  "kya_flags": [],
  "solana_tx": null,
  "solana_explorer": null
}

Response — REQUIRE_APPROVAL

{
  "outcome": "REQUIRE_APPROVAL",
  "reason": "Transaction amount 4200 USDC exceeds auto-approval threshold of 1000 USDC",
  "rule_triggered": "require_human_review",
  "hash": "sha256:c8d5…",
  "prev_hash": "sha256:5d1e…",
  "enforcement_ms": 0.44,
  "ofac_ms": 20.1,
  "total_ms": 69,
  "audit_record_id": "7a91…",
  "approval_request_id": "e2f0…",
  "policy_found": true,
  "risk_score": 40,
  "risk_level": "medium",
  "kya_flags": ["LARGE_TRANSACTION", "APPROVAL_REQUIRED"],
  "solana_tx": null,
  "solana_explorer": null
}

Every stored decision returns hash and prev_hash, forming a hash chain. Responses also carry kernel_* fields: a Rust kernel evaluates the same action in shadow mode and reports whether it agrees. It does not decide. Repeating an identical request within 5 minutes returns the original decision with "duplicate": true instead of recording a second one.

#Outcomes

Every enforcement returns exactly one of three outcomes.

Rule codes (rule_triggered)

#Policy Language (APL)

Agent Policy Language is a deterministic, human-readable DSL for spending limits and approval thresholds. Dashboard → Limits generates the policy below; editing limits there is the supported way to change them.

policy "my-defi-agent-v2" {
  version = "2.0"

  rule auto_approve {
    when input.amount <= 500
    then APPROVE
  }

  rule require_human_review {
    when input.amount > 1000 and input.amount <= 10000
    then REQUIRE_APPROVAL
    reason "Amount requires human review"
  }

  rule daily_limit {
    when daily_total(input) > 10000
    then DENY
    reason "Daily limit of 10000 USDC exceeded"
  }

  rule allowlist_only {
    when !(input.destination in ["7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJevxJSw"])
    then DENY
    reason "Destination not on allowlist"
  }
}

#Hash Chain

Every stored decision carries its own hash and the prev_hash of the decision before it. Each hash commits to the previous one, so changing any record breaks every link after it. Stored audit records cannot be updated or deleted — the database rejects both.

Verify a record

# Public — no API key. Accepts the full hash or a >= 8-char hex prefix.
curl https://solanacompliance.com/api/verify/sha256:9f2c...

# {
#   "verified": true,
#   "hash": "sha256:9f2c…",
#   "prev_hash": "sha256:41ab…",
#   "outcome": "APPROVE",
#   "solana_tx": "4vhK…",
#   "explorer_url": "https://explorer.solana.com/tx/4vhK…?cluster=devnet",
#   …
# }

#Solana Anchoring

When anchoring is enabled, each decision's hash is written to Solana devnet through the Memo program, asynchronously after the response is returned. Devnet is a public test network: it gives an independent timestamp anyone can check, but it is not a production ledger. Mainnet anchoring is on the roadmap and not live.

#Webhooks

DENY and REQUIRE_APPROVAL decisions trigger a notification to the deployment's webhook URL (Slack incoming-webhook URLs get a formatted message; any other URL gets the JSON below). Per-operator webhook URLs are not self-serve yet — ask us to route them.

Webhook payload

// POST → the deployment's notification URL, for DENY and REQUIRE_APPROVAL
{
  "event": "REQUIRE_APPROVAL",
  "agent_slug": "my-defi-agent",
  "action": {
    "type": "transfer",
    "amount": 4200,
    "currency": "USDC",
    "destination": "7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJevxJSw"
  },
  "hash": "sha256:c8d5…",
  "audit_record_id": "7a91…",
  "dashboard_url": "https://…/app/pending",
  "timestamp": "2026-09-28T11:44:22.887Z"
}

Responding to approvals

Approve or deny pending actions in Dashboard → Pending. Approvals are resolved by a signed-in operator; there is no API-key endpoint for them yet. Your agent should hold the action until it sees the outcome there, or re-submit it later.

#Audit Trail

Every decision made with an API key is stored as an append-only audit record, visible only to members of your operator. Browse it in Dashboard → Audit, or download it while signed in.

#Compliance

MiCA Art. 68(9) requires crypto-asset service providers to keep records of every service, activity, order, and transaction for 5 years — up to 7 on regulator request — and to produce them on demand. When an autonomous agent places the order, conventional application logging does not produce a record that survives an audit. Intaglio produces one at decision time.

#Errors

Errors from /api/enforce are returned as a DENY decision, so a caller that executes only on APPROVE fails safe. Where available, code gives a stable machine-readable reason.

{
  "outcome": "DENY",
  "reason": "Rate limit exceeded — 300 requests per minute per agent",
  "rule_triggered": "rate_limit"
}

Questions not covered here? Email the team or request source access.