NEWS / DESIGN

How Gantry agents work together

Parallel work needs shared evidence, clear roles, and a way to bring decisions together.

GANTRY TEAM ·
A painted tidal estuary where branching waterways converge between pale sandbanks.
Imagined landscape · AI-generated

How can several agents move one project forward?

Agents need a shared goal, a known starting state, and clear responsibility. In Gantry Crew, Bots can work on separate candidates, ask peers questions, and return results with evidence. Responsible Bots compare those results and decide what to do next within their delegated authority.

What keeps parallel work from drifting apart?

The shared record links work to its versions, constraints, and dependencies. Discussions and objections stay attached to the work they concern. This helps the team see when two changes conflict or when a result needs to be checked again. Parallel work can then come back together before the next round begins.

What are the roles of Mentor, Core, and runners?

Mentor provides context-based review and proposals in Ledger. Crew Bots own planning and technical judgment in Crew. The shared Core checks permissions, versions, dependencies, and execution conditions. Runners carry out authorized work and preserve execution progress. These roles connect reasoning, records, and action without treating a suggestion as permission to execute.

Where do external tools and digital twins fit?

Gantry connects to the tools a project uses rather than replacing each specialist tool. Code, CAD, and simulation can contribute artifacts and evidence to the shared development state. A digital twin can help evaluate a design during development or investigate behavior during operation. A result remains tied to what was actually modeled and tested.

What does autonomous work mean here?

Bots choose and carry out next steps within an approved goal and scope, using your agent runtimes and connected tools. Their ability to act stays bounded by permissions and execution checks. A successful simulation or technical review is evidence, not proof that a physical robot is safe.

How does a decision help the next task?

The team records what it chose, the evidence behind it, and what happened afterward. Later work can use that history and challenge an earlier conclusion. The aim is a development process that carries knowledge forward while remaining able to change its mind.

Written by the Gantry team. Published November 2, 2026.

Further reading: Gantry on GitHub · Ledger and Crew responsibilities. For setup and current supported behavior, use the repository documentation.

← All articles