Question 1

What are agents?

Accepted Answer

Agents are AI workers that operate across your operational data. Each agent has a specific domain (finance, operations, support, implementation) and takes actions like auditing invoices, selecting carriers, opening exceptions, or drafting a customer response. Each one runs at the autonomy level you set for it.

Question 2

How do agents learn about my operation?

Accepted Answer

Agents connect to your operational data across OMS, TMS, rate management, and spend management. They analyze your shipment history, contracted rates, carrier performance, and business rules to make decisions specific to your operation. Your data stays in the account it belongs to. We do not train models on your personal data; aggregated, de-identified data derived from your use of the platform may be used to improve it and to train our models, as described in our Terms of Service.

Question 3

Can I control what agents do?

Accepted Answer

Yes. Set the autonomy level per agent and per domain, from reviewing every action to letting the agent run and report. Thresholds decide what it may act on by itself, anything above them waits for a person, and the kill switch stops an agent at any time. You can change any of it whenever you want.

Question 4

How does graduated autonomy work?

Accepted Answer

The engine runs six levels, manual at one end and autonomous at the other. Draft an agent and simulate it against your own data before it touches anything, then publish it. Set the level per agent and per domain, so the Finance agent can file claims independently while the Operations agent still asks before it books. Every decision, tool call, and outcome is logged on a reviewable timeline.

Question 5

Which models do the agents run on?

Accepted Answer

Claude and OpenAI models today, with Google, Meta, Mistral, Groq, Hugging Face, and open-weight models served through Ollama on the full platform. Query types route to different models by cost and capability, and a local model picks up the work when a hosted provider rate-limits us. On Enterprise plans you can bring your own provider credentials and run agents on your own model. The AI models page lists every model and how each is served.

Question 6

Can agents use tools outside CommerceShip?

Accepted Answer

Through the same connections the rest of the platform runs on, so agents read and write the records your systems already share with us. 300+ integrations cover carriers, warehouses, stores, and ERPs out of the box, and on the full platform the visual workflow builder reaches 1,868+ apps, including Slack, Microsoft Teams, your ERP, and your data warehouse. On the full platform, webhooks push events to your endpoint, and a workflow routes what an agent finds to the place your team already works. MCP runs the other way: AI tools you already use, such as Claude Desktop or Cursor, connect straight to your CommerceShip data. Agents you run elsewhere can do the same over A2A: we publish an agent card, and the JSON-RPC endpoint it advertises answers inventory checks, order status, and shipment tracking. Inside the platform the agents draw on 80+ tools across the supply chain.

Question 7

What happens when an agent gets something wrong?

Accepted Answer

Every decision carries its reasoning and a confidence score, and every action is recorded on a timeline you can review, so you can see what the agent did and why. A hallucination check runs on each answer before it reaches you, regex and semantic, alongside a safety pipeline that screens for PII, profanity, and prompt injection. Actions inside the platform can be rolled back, including at full autonomy. For anything already sent outside it, a filed claim or a purchased label, the kill switch stops the agent and the timeline shows you what to follow up on. If the pattern is wrong at the source, lower the autonomy level for that agent while you correct the rule or the data it read.

Question 8

Which agents are available?

Accepted Answer

Finance Agent (invoice audit, reconciliation, rate management, claims), Operations Agent (carrier selection, label generation, tracking, warehouse and order flow, exceptions), Support Agent (data queries, knowledge base search, response drafting), and Implementation Agent (systems mapping, integration build and verification, self-healing repair). Which ones you run is scoped with you during onboarding.

Question 9

Which plans include agents?

Accepted Answer

Agents are on every plan. Autonomous agents and bring-your-own model are on Enterprise.

Hand the work to an agent and set the autonomy

Agents on CommerceShip

An agent at work

Exceptions arrive in one queue

Each decision shows which rule applied

You set the autonomy level per agent

Give an agent a job and set the level

Agent FAQs