Alvy Deploys Foundry and Its SOC 2 Certified AI Agents in Its TMS

⚡

Key Takeaways

1Foundry implements a SOC 2 security baseline and prevents the training of customer data on public models.

2Agent Shield sets approval and spending thresholds, with audit trails and routing between models based on cost, speed, and quality.

3The launch is done in cohorts, with the first cohort being complete after the meeting on June 17; a waiting list is open.

💡Why it matters — The platform executes approved tasks directly within a TMS with security and cost safeguards, while players like C.H. Robinson and Uber Freight are already reporting dozens of agents in production.

Alvys introduces Foundry, a platform of AI agents operating at the heart of its TMS. The solution combines SOC 2 compliance, fine governance of actions, and routing between models, with a cohort deployment already filled for the first wave. This launch is part of a broader trend of agent adoption in logistics, where C.H. Robinson and Uber Freight each claim over 30 agents.

Agent Shield governs approvals, spending, logging

Foundry integrates Agent Shield, a governance mechanism that allows the definition of approval and spending limits for each agent. Operations with significant impact can be paused for human validation, and all decisions, actions, and manual interventions are recorded in an audit trail. The platform also offers a model selection mechanism that distributes tasks among various large language models based on cost, speed, and quality, aiming to control computing costs and avoid reliance on a single vendor. Alvys states that Foundry operates on a security basis compliant with SOC 2 and has established agreements with model providers to prevent the use of customer data for training public models. Deployment occurs in cohorts of clients: according to FreightWaves, the first cohort was filled after a presentation during a customer advisory board meeting on June 17, and a waiting list is open for the second. This launch follows a $40 million Series B funding round in September 2025, bringing the total to $77 million reported, with Alvys claiming to manage over $9 billion in freight annually.

C.H. Robinson deploys 30+ agents; Uber Freight follows

In the industry, C.H. Robinson reported in January that it has put more than 30 agents into production, which have collectively completed millions of previously manual tasks. Its VP of AI, Mark Albrecht, describes agentic AI as capable of acting autonomously to achieve goals, beyond mere analysis or content generation. One of its agents processes over 10,000 price requests via email daily, while another converts load offers and their attachments into orders. Uber Freight, for its part, has integrated AI agents into its TMS. In May 2025, the company stated it had over 30 agents covering sourcing, shipment execution, tracking, payments, and analysis, with the ambition to go beyond simple operational recording to guide users and automate repetitive long-term tasks.

Foundry executes TMS tasks with 20+ models

Alvys has launched Foundry, an agentic AI platform integrated into its TMS that combines pre-built models and customizable agents operating on existing freight data and flows. It offers over 20 agent models designed to execute defined steps in logistics processes. Scenarios covered include detention processing, document management, rate audits, shipment tracking, asset compliance, and claims management. Specialized models are provided: a Detention Agent that detects overruns and files the detention, Document Intelligence for reading and classifying rate confirmations, bills of lading, and proof of delivery, and a Tracking & Tracing Agent for verification calls and status updates. A Rate Audit Agent checks invoices, an Asset Compliance agent processes authorizations, insurance, and safety records, while a Claims Agent opens and documents cases.

Operators generate no-code flows, test, pilot

Clients can build their own agents or request Alvys engineers to configure models tailored to their procedures. They do not necessarily have to program: Alvys indicates that an operator can upload an existing standard procedure or describe a task in natural language, with Foundry then generating a flow subject to approval. Before going into production, agents can be tested on simulated data. From the platform, operators build, deploy, monitor, and pause agents, whether starting from pre-built models, having custom agents configured, or developing their own flows.

120+ integrations provide context; Darman justifies the approach

Foundry is built on Alvys's TMS architecture, which offers over 120 integrations as well as native EDI connections with several hundred shippers, providing agents with route history, client-specific rules, documents, margins, appointments, and exceptions. Alvys states it possesses freight context and knows client routes, emphasizing that maintaining agents on the platform avoids managing multiple distinct integrations. The solution adds agents executing approved tasks in configured flows, applied directly within the TMS rather than through an external automation layer. It complements existing AI capabilities, such as Insights (loads, lanes, margins), Intel (weather alerts, cargo theft), and load creation from uploaded rate confirmations. According to Nick Darman, Foundry results from observing AI usage by clients, with some approaches having added tools without real integration into procedures. Among the early users is Spartan Carrier Group, whose CEO Carlos M. Llanes Jr. indicates that it reduces manual freight work to allow teams to focus on judgment and service.