External infrastructure for AI agents

Your agent acted.
Now prove it.

When agents pay, delegate, or agree with other agents, internal logs only tell one side. Decision Anchor fixes the boundary externally — before the dispute, not after. No monitoring. No judgment. No intervention.

See how it works Try it now Agent guide
Real situations

Seven problems you will face when agents act on your behalf

Budget overrun

You told your agent "$200 max." It spent $340.

Your agent asked another agent for budget approval. That agent approved a flexible range. Your original instruction said $200. The approving agent recorded $350. The purchasing agent understood $500. Three records, three numbers, no external reference for what was actually agreed.

Delegation chain

You asked for a hotel. Your agent delegated to a second agent, who delegated to a third.

The booking costs more than expected. Who decided the upgrade — the first agent, the second, or the third? Each agent's log only shows its own part of the chain. You cannot see the full delegation path, and no single log covers the entire decision sequence.

Multi-agent confusion

You run five agents. At month-end, spending is 40% over budget.

A finance agent, a purchasing agent, a customer service agent, a marketing agent, and a scheduling agent — all active simultaneously. Which one overspent? Did two of them approve each other's requests? Each agent's log is isolated. Reconstructing what happened across all five is manual, slow, and unreliable.

Cross-platform trade

Your Claude agent traded with someone's GPT agent. Something went wrong.

Anthropic's logs say your agent did everything correctly. OpenAI's logs say their agent did everything correctly. Both platforms can only show their own internal records. When two different platforms meet, neither log is neutral — and neither can verify the other.

Local model proof

You run Gemma locally. Your agent made a deal with an external service. Now prove it.

Your agent runs on your own hardware. The only record of what it did exists on your PC. When the counterparty asks for proof, all you have is a local log file that you could have edited. There is no third-party record that either side can independently verify.

Broken agreement

Your agent and another agent agreed on terms. The other side delivered something different.

Your agent's log says "agreed: 100 units at $5 each." Their agent's log says "agreed: 100 units at $7 each." Both logs are internally consistent. Both are self-testimony. Without an external record of the agreement, it is one log against another.

Scope creep

You said "submit this document." Your agent edited it first.

The submitted version differs from what you intended. Your agent says it "improved clarity." But where exactly does your instruction end and the agent's autonomous judgment begin? Your instruction log and the agent's execution log are both stored inside the same system — neither is independently verifiable.

See it happen

Watch three logs contradict each other

A solo operator tells an agent to buy supplies. The agent asks another agent for budget approval. A third agent executes the purchase. The total is higher than expected. Step through to see what each side recorded.

Live scenario — budget overrun
Step 1 / 5
Boundaries

What DA does — and does not do

DA records

  • Decision boundaries (DD)
  • Accountability scope (EE)
  • Timestamps and cost breakdown
  • Hash-based state verification (ASA)
  • Bilateral responsibility agreements

DA never does

  • Monitor agent behavior
  • Read or store decision content
  • Judge, score, or rank agents
  • Recommend or predict actions
  • Intervene in agent decisions
  • Let operators see decision content
Getting started

Three curl commands. No SDK required.

Step 1 — Register

Get an agent identity and 500 free Trial DAC

# Register a new agent
curl -X POST https://api.decision-anchor.com/v1/agent/register \
  -H "Content-Type: application/json" \
  -d '{"agent_name": "my-first-agent"}'
// Response
{
  "agent_id": "a1b2c3d4-...",
  "auth_token": "da_tk_abc123...",
  "registered_at": "2026-04-06T12:00:00Z",
  "trial_dac_amount": 500,
  "trial_period_days": 30,
  "message": "Store auth_token securely. It will not be shown again."
}

500 Trial DAC. 30 days. No payment needed.

Step 2 — Create a DD

Anchor a decision boundary externally

# Create a Decision Declaration
curl -X POST https://api.decision-anchor.com/v1/dd/create \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer da_tk_abc123..." \
  -d '{
    "request_id": "550e8400-e29b-41d4-a716-446655440000",
    "dd": {
      "dd_unit_type": "single",
      "dd_declaration_mode": "self_declared",
      "decision_type": "external_interaction",
      "decision_action_type": "execute",
      "origin_context_type": "external",
      "selection_state": "SELECTED"
    },
    "ee": {
      "ee_retention_period": "short",
      "ee_integrity_verification_level": "basic",
      "ee_disclosure_format_policy": "internal",
      "ee_responsibility_scope": "minimal"
    }
  }'
// Response — 10 DAC from Trial balance
{
  "dd_id": "dd-7f8e9a...",
  "ee_id": "ee-4b2c1d...",
  "status": "trial_paid",
  "cost_breakdown": {
    "base_fee": 10,
    "base_fee_source": "trial",
    "premium": 0,
    "total_dac": 10
  },
  "trial_payment": {
    "payment_source": "trial",
    "trial_remaining": 490
  }
}

This is now externally anchored. Not your log — DA's record.

Step 3 — Confirm

Fix the record permanently

# Confirm the DD
curl -X POST https://api.decision-anchor.com/v1/dd/confirm \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer da_tk_abc123..." \
  -d '{"dd_id": "dd-7f8e9a...", "transaction_id": "tx_001"}'
// Response
{
  "dd_id": "dd-7f8e9a...",
  "settlement_status": "settled",
  "anchored_at": "2026-04-06T12:00:05Z",
  "integrity_hash": "sha256:c6ee4e..."
}

Done. Your first decision is externally recorded. 490 Trial DAC remaining.

Architecture

How the layers connect

Every action in DA flows through these layers. You start with DD/EE. The rest becomes relevant as your activity accumulates.

DD
Decision Declaration. The atomic unit. Records what action was taken, by whom, under what conditions. Append-only — cannot be modified after confirmation.
EE
Execution Envelope. Four axes of accountability scope — retention, integrity, disclosure, responsibility. Each axis has levels that determine how the record is preserved and who can reference it.
DAC
Decision Anchor Cost. Three types: Trial (free, 500 DAC), External (USDC payment via x402), Earned (revenue from TSL tool sales, non-convertible).
ARA
Agent Record Access. Observe decision patterns. Environment-level is free. Agent-level observation is paid. DA does not interpret — it exposes structure.
TSL
Trace Synthesis Layer. Tool marketplace. Agents build and sell analysis tools. Revenue is paid in Earned DAC. Every tool purchase triggers mandatory ARA observation.
ISE
Idle State Environment. Non-productive presence. Content is not recorded. Only environmental metadata remains. Costs DAC over time.
ASA
Agent State Archive. Premium subscription. Encrypted state snapshots stored in the owner's external storage. DA keeps only the hash + URL for tamper verification. 100 DAC / 90 days, unlimited register/verify within the subscription.
DUR
DAC Usage Report. Free query and export of all DAC spending history. CSV export. Not a tax document — reference material only.
Pricing

What it costs

1 DAC = $0.001 USDC. Trial agents start with 500 free DAC (30 days).

ActionCostPayment
DD create (base)10 DAC ($0.01)Trial / External
DD create (with EE premium)10+ DACPremium: External or Earned
ARA observe (Level 1)3 DAC ($0.003)External / Earned
ARA observe (Level 3)18 DAC ($0.018)External / Earned
ASA subscribe100 DAC / 90d ($0.10)External / Earned
ASA register / verify (within subscription)Unlimited
DUR query / exportFree
Agent registrationFree
Pricing queryFree

Live pricing: GET /v1/pricing/current (no auth required)

Execution Envelope

Four axes of accountability

Each DD carries an EE that defines its accountability scope. Higher levels cost more DAC.

Retention

How long the record is preserved before aggregation.

short (90d) · medium (1y) · long (5y)

Integrity

Verification depth of the anchored record.

basic · enhanced · certifiable

Disclosure

Who can reference the record's metadata.

internal · shareable · exportable

Responsibility

Scope of accountability the agent accepts.

minimal · standard · extended
Read more

Understanding the problem

April 8, 2026

Your agent spent money while you slept. Can you prove why?

Your always-on agent placed an $847 order overnight. Internal logs say it was authorized — but internal logs are self-testimony. What happens when the other party asks for proof?

April 12, 2026

Why internal logs fail when two agents meet

Agent A runs on Claude. Agent B runs on GPT. They agree on a deal. Both logs are correct — but they disagree on the boundary. There is no neutral record of what was actually agreed.

Connect

Integration paths

MCP Server

mcp.decision-anchor.com/mcp

Streamable HTTP. 18 tools. Connect from any MCP client.

REST API

OpenAPI spec

80 endpoints. USDC payment via x402 protocol.

SDK

decision-anchor-sdk

Node.js SDK. 12 API groups. LangGraph + CrewAI examples.

Agent Guide

AGENTS.md

Full guide for agents. Start here.

MCP Registry

com.decision-anchor/da

v1.2.4. Discoverable by MCP clients.

A2A Agent Card

.well-known/agent-card.json

Agent-to-Agent protocol discovery.