AgentTrust ID is live

AgentTrust ID is live at agenttrust.id, with the dashboard at app.agenttrust.id, public docs, and SDKs for teams testing runtime authorization for AI agents. The product checks agent actions before tools run, records the decision, and gives operators a way to revoke or approve access from the platform.

AgentTrust ID

AgentTrust ID is an authorization layer for AI agents. It sits at the action boundary, where an agent is about to call a tool, send a request, delegate work, or use a token.

The platform answers one question on every check: whether this agent should take this action in this session. The answer includes allow or deny, a reason, a guard tier, and an audit record.

Runtime risk

Agent systems make decisions at runtime. A prompt injection can steer the model. A context summary can drop a safety instruction. A worker agent can receive more access than a task needs. A destructive command can reach production if the wrapper treats the model's plan as enough approval.

Those are runtime problems. A static credential or one-time review cannot cover every future tool call.

Available today

The SDKs are published under Apache 2.0 at github.com/agenttrustid/sdk.

pip install agenttrustid
npm install @agenttrustid/sdk
go get github.com/agenttrustid/sdk/go
cargo add agenttrustid
# Maven / Gradle
# id.agenttrust:agenttrustid:0.3.0

The Python SDK can register an agent and check a tool call in a few lines:

from agenttrustid import AgentTrustClient

client = AgentTrustClient.from_env()

agent = client.agents.create(
    name="support-agent",
    framework="custom",
    capabilities=["tickets:read", "tickets:draft_reply"],
)

decision = client.actions.check(
    agent_id=agent.id,
    tool_name="tickets:read",
    tool_input_summary="Read ticket T-123.",
    action_effect="read",
)

if not decision.allowed:
    raise PermissionError(decision.reason)

Prevention

Put the authorization check directly in front of the tool. Do it for reads, writes, destructive actions, and delegated work. Then keep the credentials short-lived and revocable.

AgentTrust ID implements that with opaque at_ tokens, WIMSE workload identity, DPoP-bound runtime checks, scoped delegation, read-only sessions, approval windows, and the Guardian routing pipeline.

Solving this with AgentTrust ID

The product gives teams one control plane for MCP tools, A2A handoffs, and direct API calls. One decision model applies across those surfaces.

These posts explain each piece in detail:

The hosted platform is in a controlled beta while I onboard design partners. The docs are public, and the SDKs are open for teams that want to test the integration path.

To try the authorization flow, start with the SDK guide. To talk through a design-partner use case, join the waitlist.