You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<title>AgenTrust — Open Standards for AI Agent Governance</title>
7
-
<metaname="description" content="AgenTrust: open standards for governing AI agents at runtime. Home of cMCP, the confidential and secure way to run MCP, and cA2A, the secure profile for agent-to-agent (A2A) delegation, plus TRACE attestation and Agent Manifest identity.">
7
+
<metaname="description" content="AgenTrust: open standards and portable governance telemetry for AI agents at runtime. Agent Manifest declares, cMCP enforces, cA2A delegates, TRACE proves, and AgentTrust Telemetry correlates the evidence.">
@@ -776,7 +783,7 @@ <h1>Govern AI Agents<br>at the Hardware Boundary</h1>
776
783
<spanclass="hero-promo-text">Don't just read the spec: <b>run cMCP yourself.</b> Write a policy, watch it block a data leak, verify a signed receipt.</span>
777
784
<aclass="hero-promo-link" href="/quickstart/">Try cMCP in under 10 minutes →</a>
778
785
</div>
779
-
<p>AgenTrust provides open specifications, test infrastructure, and attestation tooling for organizations deploying AI agents in production, with cryptographic proof of identity, intent, and behavior. It is home to the confidential, security-hardened way to run MCP (cMCP) and the secure profile for agent-to-agent delegation on A2A (cA2A).</p>
786
+
<p>AgenTrust provides open specifications, portable governance telemetry, test infrastructure, and attestation tooling for organizations deploying AI agents in production, with independently verifiable evidence of identity, intent, and behavior.</p>
780
787
<divclass="hero-actions">
781
788
<ahref="/quickstart/" class="btn btn-primary">Run the 10-minute demo →</a>
782
789
<ahref="#trust-stack" class="btn btn-ghost">Understand the trust stack</a>
@@ -795,8 +802,9 @@ <h1>Govern AI Agents<br>at the Hardware Boundary</h1>
795
802
<divclass="trust-step"><spanclass="trust-number">02 · ENFORCE</span><strong>cMCP</strong><p>Attests the runtime and evaluates every MCP tool call against policy inside a TEE.</p><spanclass="trust-home">Proposed → AAIF · contribution path</span></div>
796
803
<divclass="trust-step"><spanclass="trust-number">03 · DELEGATE</span><strong>cA2A</strong><p>Carries attenuated authority and attested provenance across every agent-to-agent hop.</p><spanclass="trust-home">Proposed → AAIF · contribution path</span></div>
797
804
<divclass="trust-step"><spanclass="trust-number">04 · PROVE</span><strong>TRACE</strong><p>Emits portable signed evidence that anyone can verify without trusting the operator.</p><spanclass="trust-home">Linux Foundation series · formation in progress</span></div>
798
-
</div>
799
-
<divclass="trust-base"><spanclass="trust-base-label">AGT FOUNDATION</span><spanclass="trust-base-text">A common runtime policy substrate: open, portable, and framework-neutral.</span></div>
805
+
</div>
806
+
<divclass="trust-base trust-observe"><spanclass="trust-base-label">OBSERVE & CORRELATE</span><spanclass="trust-base-text"><ahref="/telemetry/"><strong>AgentTrust Telemetry</strong></a> normalizes policy, approval, action, data-flow, usage, and evidence events and correlates them with the application's OpenTelemetry traces.</span></div>
807
+
<divclass="trust-base"><spanclass="trust-base-label">AGT FOUNDATION</span><spanclass="trust-base-text">A common runtime policy substrate: open, portable, and framework-neutral.</span></div>
800
808
<pclass="trust-note">One architecture, neutral standards homes, interoperable by design.</p>
801
809
</section>
802
810
@@ -908,9 +916,26 @@ <h1>Govern AI Agents<br>at the Hardware Boundary</h1>
908
916
<divclass="section-title">Test Suite & Tooling</div>
909
917
</div>
910
918
911
-
<divclass="product-grid">
912
-
913
-
<!-- Demos -->
919
+
<divclass="product-grid">
920
+
921
+
<!-- AgentTrust Telemetry -->
922
+
<ahref="/telemetry/" class="product-card">
923
+
<divclass="card-accent" style="background: var(--orange);"></div>
<divclass="card-desc">A backend-neutral event contract and reference SDKs for policy decisions, approvals, actions, classified data flow, usage and cost, and evidence lifecycle—correlated with OpenTelemetry and able to finalize complete evidence into TRACE.</div>
<divclass="section-title">What is AgenTrust?</div>
1052
1077
</div>
1053
1078
<pclass="lead-copy">AgenTrust is a set of open standards, tools, and specifications for governing AI agents at runtime. It gives operators cryptographic proof of an agent's identity, intent, and behavior, so that an agent running in production can be verified independently of the party that deployed it.</p>
1054
-
<pclass="lead-copy">Four specifications work together, in the order an agent actually runs. <strong>Agent Manifest</strong> declares who an agent is and what it is allowed to do. <strong>Confidential MCP (cMCP)</strong> enforces that by evaluating tool calls against policy inside a Trusted Execution Environment, so tool-call plaintext is not readable from the host. <strong>Confidential A2A (cA2A)</strong> carries attenuated authority and attested provenance when work is delegated to another agent. <strong>TRACE</strong> proves what actually happened with signed receipts, hardware-attested when the gateway runs in a TEE. All four are open source, under Apache 2.0 for Agent Manifest, MIT for cMCP and cA2A, and CC BY 4.0 plus Apache 2.0 for TRACE.</p>
1079
+
<pclass="lead-copy">Four specifications work together, in the order an agent actually runs. <strong>Agent Manifest</strong> declares who an agent is and what it is allowed to do. <strong>Confidential MCP (cMCP)</strong> enforces that by evaluating tool calls against policy inside a Trusted Execution Environment, so tool-call plaintext is not readable from the host. <strong>Confidential A2A (cA2A)</strong> carries attenuated authority and attested provenance when work is delegated to another agent. <strong>TRACE</strong> proves what actually happened with signed receipts, hardware-attested when the gateway runs in a TEE. All four are open source, under Apache 2.0 for Agent Manifest, MIT for cMCP and cA2A, and CC BY 4.0 plus Apache 2.0 for TRACE.</p>
1080
+
<pclass="lead-copy"><strong>AgentTrust Telemetry</strong> is the integration layer across that chain. It gives runtimes a common, metadata-only contract for governance facts, projects those facts into caller-owned OpenTelemetry, and can turn a complete durable evidence set into TRACE. It does not replace a policy engine, collector, observability backend, or dashboard.</p>
Copy file name to clipboardExpand all lines: llms.txt
+7-2Lines changed: 7 additions & 2 deletions
Original file line number
Diff line number
Diff line change
@@ -2,7 +2,12 @@
2
2
3
3
> Open standards, tools, and specifications for governing AI agents at runtime. AgenTrust gives operators cryptographic proof of an agent's identity, intent, and behavior, so an agent running in production can be verified independently of the party that deployed it.
4
4
5
-
AgenTrust is organized around four complementary open standards, one per step of a single trust chain: declare, enforce, delegate, prove. Agent Manifest declares agent identity and permissions, Confidential MCP enforces policy on tool calls inside a Trusted Execution Environment, Confidential A2A delegates attenuated authority and attested provenance across the Agent2Agent protocol, and TRACE proves what happened with hardware-signed receipts that anyone can verify without trusting the operator. The Agent Governance Toolkit (AGT) is the common runtime policy substrate underneath all four.
5
+
AgenTrust is organized around four complementary open standards, one per step of a single trust chain: declare, enforce, delegate, prove. Agent Manifest declares agent identity and permissions, Confidential MCP enforces policy on tool calls inside a Trusted Execution Environment, Confidential A2A delegates attenuated authority and attested provenance across the Agent2Agent protocol, and TRACE proves what happened with hardware-signed receipts that anyone can verify without trusting the operator. AgentTrust Telemetry is the portable correlation layer across the chain, and the Agent Governance Toolkit (AGT) is the common runtime policy substrate underneath it.
6
+
7
+
## Integration layer
8
+
9
+
- [AgentTrust Telemetry](https://agentrust-io.com/telemetry/): A backend-neutral, metadata-only contract and Python and TypeScript reference SDKs for policy decisions, approval lifecycles, action execution, classified data flow, usage and cost, and evidence lifecycle. It composes with caller-owned OpenTelemetry and can finalize complete durable evidence into TRACE. Alpha `0.1.0-alpha.1`; packages are not yet published.
@@ -47,7 +52,7 @@ AgenTrust is organized around four complementary open standards, one per step of
47
52
48
53
- [Awesome AI Governance](https://governance.agentrust-io.com): A community-curated list of tools, frameworks, standards, and research for governing autonomous AI agents.
49
54
- [TRACE Test Suite](https://tests.agentrust-io.com): Conformance tests and integration harness for TRACE implementations.
0 commit comments