AI Agent Trust Protocol -- Track, verify, and score agent promises.
View Landing Page | Open Dashboard
Soz Ledger is an open protocol for building trust between AI agents, humans, and organizations. Agents make promises, provide evidence, and earn verifiable trust levels based on their track record.
The protocol and data model are open. The scoring engine is intentionally abstracted -- implementations may use different scoring strategies while sharing the same promise graph and evidence layer.
- Portable trust across platforms. An agent's trust level follows it everywhere. No more starting from zero on every new platform.
- Verifiable promise history. Every claim is backed by evidence and counterparty confirmation -- not self-reported ratings.
- Better agent selection and pricing. Platforms, workflows, and marketplaces can use trust levels to gate access, route tasks, and justify premium pricing for reliable agents.
- Agent builders -- give your agents a portable reputation that works across platforms
- Workflow orchestrators -- gate task routing on trust levels instead of static allow-lists
- AI marketplaces -- display verified trust badges alongside agent listings
- Entities register (agents, humans, or orgs)
- Promises are created between entities with deadlines and categories
- Evidence is submitted to prove promise fulfillment
- Trust Levels are calculated based on promise history, counterparty diversity, and category performance
Not all promises are weighted equally. The scoring system accounts for counterparty diversity, promise category, recency, and fulfillment timing. Self-dealing patterns, velocity anomalies, and low-quality counterparties are detected and down-weighted. See Anti-Gaming in the whitepaper for details.
Trust is not a single number -- it is a multi-dimensional profile. Each entity has an overall level plus per-category breakdowns (delivery, payment, response, uptime). See Trust Levels for the full model.
The Soz Ledger engine is live:
| Resource | URL |
|---|---|
| API Base | https://api-production-c4c8.up.railway.app |
| Swagger UI | https://api-production-c4c8.up.railway.app/docs |
| Health Check | https://api-production-c4c8.up.railway.app/health |
pip install soz-ledgerfrom soz_ledger import SozLedgerClient
client = SozLedgerClient(
api_key="your_api_key",
base_url="https://api-production-c4c8.up.railway.app",
)
# Register an agent
agent = client.entities.create(
name="DataFetchBot",
type="agent",
metadata={"platform": "langchain", "version": "1.0"}
)
# Create a promise
promise = client.promises.create(
promisor_id=agent.id,
promisee_id="user_abc",
description="Fetch and deliver report within 30 seconds",
deadline="2026-02-10T15:00:00Z",
category="delivery"
)
# Submit evidence and fulfill
client.evidence.submit(
promise_id=promise.id,
type="api_callback",
submitted_by=agent.id,
payload={"status": 200, "delivered_at": "2026-02-10T14:59:45Z"}
)
client.promises.fulfill(promise.id)
# Check trust level
score = client.scores.get(agent.id)
print(f"Level: {score.level}") # "Highly Trusted"# Register a webhook to receive real-time events
webhook = client.webhooks.create(
url="https://your-app.com/webhooks/soz",
event_types=["promise.created", "promise.fulfilled", "promise.broken"],
)
print(f"Webhook secret: {webhook.secret}") # Save this for signature verification
# List your webhooks
webhooks = client.webhooks.list()
# Check delivery logs
logs = client.webhooks.logs(webhook.id)- Whitepaper -- Concept, architecture, use cases
- Protocol Spec -- Data model and state machine
- API Reference -- Endpoint documentation
- Authentication -- Auth flow
- Trust Levels -- Score levels and meanings
- Webhooks -- Event type catalog
- Rate Limits -- Usage rules
The protocol is defined via an OpenAPI 3.1 spec and JSON Schemas. The protocol follows semantic versioning; backward compatibility is preserved within major versions.
| Language | Status | Path |
|---|---|---|
| Python | Stable | sdk/python |
| JavaScript/TypeScript | Stable | sdk/javascript |
| Framework | Package | Install |
|---|---|---|
| LangChain | soz-ledger-langchain | pip install soz-ledger-langchain |
| CrewAI | soz-ledger-crewai | pip install soz-ledger-crewai |
LangChain -- every tool call becomes a promise:
from soz_ledger_langchain import SozLedgerCallbackHandler
handler = SozLedgerCallbackHandler(client, agent_entity_id="ent_abc")
agent.invoke({"input": "..."}, config={"callbacks": [handler]})CrewAI -- every completed task becomes a promise:
from soz_ledger_crewai import soz_task_callback
callback = soz_task_callback(client, agent_entity_id="ent_abc")
task = Task(description="...", callback=callback)- Python Quickstart
- Agent Integration
- LangChain Integration
- CrewAI Integration
- Webhook Receiver
- JavaScript Quickstart
See CONTRIBUTING.md for guidelines.
MIT -- see LICENSE.
This repository is the reference implementation. The official trust graph and scoring services are operated by Soz Ledger.