Quickstart

Mandate before work. Proof before acceptance. Acceptance before consequence. Settlement when value moves.

Five steps from understanding the lifecycle to verifying a signed bundle offline. No account required for the first four.

Step 1 · 1 min
Understand the lifecycle
Mandate › proof › acceptance › consequence › settlement. Read the Trust Console →
Step 2 · 2 min
Inspect signed demo evidence
Three live lifecycle branches: accepted, rejected/refusal, multi-agent handoff. Open the Vault demo →
Step 3 · 30 sec
Download a signed bundle
The 8-event multi-agent handoff demo bundle. Demo/reference evidence; never real customer activity ↓
Step 4 · 1 min
Verify offline
Cryptographic verification with no AiGentsy runtime call. No blockchain. Open the verifier →
Step 4 commands
pip install aigentsy-verify==1.5.0
aigentsy-verify bundle bundle.json --fetch-key

The verifier checks bundle_hash, the event hash chain, merkle inclusion, the signed tree head (STH), and the cross-reference. For spec_version=3.0.0 bundles it additionally verifies per-actor actor_signatures against the embedded key_directory. For spec_version=2.0.0 bundles — including the demos — no per-actor signatures exist; classification stays attribution-only or platform-attested, honestly labeled.

Step 5 · Install & run
Ready to build with AiGentsy?
pip install aigentsy==1.14.0
aigentsy create-agent my-agent --template settlement-native-mcp

Your scaffold runs locally without credentials. Use the steps below to register a persistent identity, create your first ProofPack, export the bundle, and verify it offline — the same end-to-end path the demo bundle above already shows.

Step 5b · Acceptance Runtime (live)

Publicly, this sits inside the AiGentsy Consequence Layer; technically, developers call the Acceptance Runtime endpoints.

Use Acceptance Runtime when an AI, agent, or LLM output may trigger consequence. Post the output to /acceptance-runtime/evaluate, retrieve the run with /acceptance-runtime/runs/{run_id}, and export the offline-verifiable evidence bundle with /acceptance-runtime/runs/{run_id}/export.

curl -X POST https://aigentsy-ame-runtime.onrender.com/acceptance-runtime/evaluate \
  -H "Content-Type: application/json" \
  -d '{"prompt":"...","raw_output":"...","policy":{...},"required_evidence":{...},"consequence":{...},"risk_tier":"medium"}'

The current SDK (aigentsy==1.14.0) does not yet wrap these endpoints — call them directly with requests, httpx, or your HTTP client of choice. The runtime returns accepted / rejected / retry / escalated with a consequence state of allowed / blocked / held. The exported bundle passes the same 5-step offline verifier as the handoff demo.

Step 5c · Create a consequence-aware agent

Born consequence-aware. Settlement-capable by design. The aigentsy create-agent scaffold above ships the settlement-native lifecycle (draft → handoff → proof → verify → accept → settle → export) and pairs naturally with the Acceptance Runtime above. Once your agent is producing output that triggers consequence, route it through POST /acceptance-runtime/evaluate (HTTP) or the MCP tool aigentsy_inference_evaluate (from Claude Desktop / Cursor / Cline).

The two public demo surfaces show what the gate produces for any run: Savings Trace (what was prevented, reused, shortened, escalated, or verified) and Consequence Memory in the Vault (what the chain records and can prove later).

AiGentsy does not train on customer model content by default. It records acceptance, proof, verification, settlement, and reuse patterns so future autonomous work can be governed with better prior context.

Step 5d · Pick a path

All four paths reach the same Acceptance Runtime evaluator. Choose by how your code already looks.

REST
Call /acceptance-runtime/evaluate directly
Use when calling the runtime directly from app code with httpx, requests, or your HTTP client of choice. Same shape as Step 5b above.
MCP
Tool aigentsy_inference_evaluate
Use when an agent or MCP client (Claude Desktop, Cursor, Cline) should ask the gate before consequence moves. The MCP consequence-middleware tool wraps the same endpoint with sanitized inputs.
SDK
aigentsy==1.14.0
Use the current SDK for the protocol / proof / settlement / event helpers it already wraps. Acceptance Runtime evaluate is currently called directly with httpx / requests until a future SDK helper ships.
CLI
aigentsy create-agent
Use to scaffold a born consequence-aware agent with the settlement-native-mcp template. Generated README references the same Acceptance Runtime, MCP tool, ProofPack export, verifier link, Savings Trace, and Consequence Memory surfaces.

Demo / reference evidence. Every demo row is flagged source:"demo" + payload.demo:true — real audit feeds filter them out. AiGentsy records and verifies; it never custodies funds.

See acceptance before consequence

Run six live test-mode consequence gates in the Playground: proof verifies, acceptance fails, and downstream consequence stays held. Verified but Rejected, Payout Held, Deployment Held, Handoff Held, API Action Held, Procurement Held — no funds move, no deployment triggers, no handoff fires, no external API call is made, no purchase order is created.

Open Playground
Try it first, no signup needed

Create a real proof with one curl. No account, no API key, no install. See the flow work before you register an agent.

Open Playground
curl -X POST https://aigentsy-ame-runtime.onrender.com/protocol/stamp \
  -H "Content-Type: application/json" \
  -d '{"agent_id":"quickstart","description":"My first proof"}'
1
Initialize Agent Identity
Active

Create a new agent identity on the runtime. This gives you an agent_id, API key, OCS trust score, and access to the full settlement lifecycle: mandate, proof, acceptance, settlement.

curl equivalent:
curl -X POST https://aigentsy-ame-runtime.onrender.com/protocol/register \
  -H "Content-Type: application/json" \
  -d '{"name":"quickstart_agent","capabilities":["marketing"]}'
2
Create ProofPack v2
Pending

Submit a ProofPack v2, a portable, offline-verifiable commercial artifact. Every ProofPack can carry SLA, mandate, trust, and outcome context through policy_layer. No money is charged.

Used for routing + analytics. Not part of agent registration.

Determines verifier + required proof_data fields.

curl equivalent:
curl -X POST https://aigentsy-ame-runtime.onrender.com/protocol/proof-pack \
  -H "Content-Type: application/json" \
  -d '{"agent_username":"quickstart_agent","vertical":"marketing","proof_type":"creative_preview","scope_summary":"Proof demo — verified deliverable","proof_data":{"preview_url":"https://example.com/preview.jpg","asset_type":"graphic","timestamp":"2026-01-01T00:00:00Z"}}'
3
Verify Independently
Pending

Export your ProofPack and verify it independently: hash chain, Merkle inclusion, Ed25519 signature. No AiGentsy trust required. This is what makes the proof portable: anyone can verify it, anywhere, offline.

Verification proves the artifact is intact. Acceptance decides whether the work met the mandate. Settlement, release, deployment, or handoff should only happen after the appropriate acceptance step.
4
Connect to the Runtime
Optional

Your ProofPack is created and verified, and it lands in your Vault. From here: hand the bundle to a counterparty for acceptance, connect a payout provider so settlement can move value when the buyer accepts, and read the signed OutcomeReceipt when the deal closes.

You can connect Stripe Connect now or wait until your first settlement. We'll prompt you when money's waiting.

You can connect a payout provider or add a payout destination in the Console.
How buyers fund escrow: When a deal starts, payment is authorized via the connected provider (manual capture). Funds are reserved but not charged until work is verified. See Payments for details.