TTRL / Home
ENGINEERING SHOWCASE / TTRL AGENT RUNTIME

An agent can propose.
Authority is earned.

Security-first execution infrastructure for AI agents: Rust trust core, Go MCP gateway, deterministic policy and signed audit evidence.

Engineering evaluation

The repository documents core engineering and verification work. Customer-environment sign-off remains open and is required per deployment. This page does not claim independent certification or production acceptance.

The problem.
The control boundary.

Model-generated requests should not automatically become privileged commands. Agent and tool integrations need an explicit point where identity, policy and approval determine what may execute.

The documented runtime places a Go MCP gateway and a Rust-authoritative policy core between agent intent and verified tool dispatch, with result checks and signed audit evidence.

  1. Agent intentUntrusted action request
  2. Identity & policyAllow, deny or require approval
  3. Controlled dispatchVerified tool channel
  4. Result & evidenceOutput checks and audit records

Rust trust core

Deterministic TAPL policy, authorisation and bounded execution controls.

Go MCP gateway

Request validation, action firewall and verified MCP tool dispatch.

Scoped capabilities

Brokered secret operations and explicit boundaries around privileged workloads.

Reviewable evidence

Signed, tamper-evident dispatch, outcome and failure records as documented in the source.

Inspect the thinking.
Evaluate the evidence.

The repository README documents an Ubuntu-targeted buyer engineering demo and exact-commit evidence. Its demo is not an independent penetration test or a production deployment.

Authoritative project: MosinShaikh505/TTRL-Agent-Runtime. Source reviewed: its README on 7 November 2026. Readiness must be re-evaluated for the selected commit and target environment.

FROM EXPLORATION TO A CLEAR NEXT STEP

Let’s shape your next system.

Discuss your requirement