Validate intent
Go MCP gateway and action firewall at the tool-request boundary.
Explore how an agent request moves through policy, human approval and an audit trail. Built around the runtime’s documented Rust trust core and Go MCP gateway architecture.
Current project: MosinShaikh505/TTRL-Agent-Runtime ↗ · Private repository; GitHub access required.
This connection uses fixed scenarios through the Rust policy bridge when its isolated API is available. It does not execute tools, issue human-approval grants or create full runtime-ledger receipts.
Live policy service is not connected. The walkthrough below is a browser simulation.
No connected evaluation has run.
Fixed sample rules. No model, network request or executable workload runs.
Select a sample request, then evaluate its illustrative policy.
Review the sample internal-report scope. This control only advances the browser demonstration.
Ready. Select a sample request.
This is an unsigned browser event list. The real runtime’s signed audit evidence must be generated and verified in an actual configured environment.
The repository describes a gateway that validates requests, a Rust core that evaluates deterministic TAPL policy and approvals, bounded tool execution, and signed audit evidence. This walkthrough explains those boundaries without impersonating a live deployment.
Go MCP gateway and action firewall at the tool-request boundary.
Deterministic policy, scoped capabilities and bound approval continuations.
Governed STDIO and Wasm brokers, with environment-dependent containment.
Signed dispatch, outcome and failure evidence in the actual runtime.
The authoritative repository is MosinShaikh505/TTRL-Agent-Runtime. Its README was reviewed on 7 November 2026 and records seven unclosed parent exit gates: independent receipt trust, combined containment and quota pressure, identity-provider acceptance, operational replication/failover, genuine hardware quotes, installer platforms and final audit/documentation.
Customer-environment sign-off remains open. This page does not claim a completed production release, independent security certification, live remote administration or real hardware attestation.
Fictional CRM, one-shot approval, local investigation, browser Ed25519 verification, journal archival and security-control models. Clearly labelled simulation; no live gateway or customer administration.
Launch the interactive buyer demo →