Governed software-development runs

Give every agent run a project, a plan, a sandbox, and a review path.

xmode is the operating layer for teams that want AI to help ship software without losing ownership. Start from an objective, route through versioned skills and actions, inspect the sandbox, and finish with evidence attached to a Change Request.

Open app Explore product Read the model

Objective Plan Skill Action Sandbox Evidence Change Request

What changed

The graph is policy. The chat is how the run happens.

Pipelines are reusable wrappers for actions, decisions, follow-ups, approvals, schedules, and event triggers. When a team runs one, xmode presents the work as a structured chat with clear choices, open-ended follow-ups, goal checks, and pause points.

Project context

Issues, projects, cycles, views, events, and Change Requests keep automation anchored to real work.

Versioned catalogs

Skills, actions, and pipelines behave like editable documents with semantic versions and latest releases.

Interactive runs

Multiple-choice decisions, clarifying questions, and goal checks become structured step outputs.

Sandbox workbench

Files, commands, logs, artifacts, diffs, and review state stay visible while the agent works.

Repository import

GitHub and GitLab connections can sync repositories, including private repos through provider auth.

Event SDKs

Node, Python, and Ruby libraries can send signed product events into the Event Inbox.

Why teams use it

Superpowers for teams, not invisible autonomy.

xmode is designed for engineering teams that want leverage. It makes agent work repeatable, inspectable, interruptible, and compatible with the review process teams already trust.

Delegate bounded work

Run implementation, dependency updates, failing-build fixes, event handling, and review workflows from project context.

Keep judgment in the loop

Pause for approvals, ask follow-up questions, revise plans, or branch the run when the objective is not clear.

Review the actual work

Inspect sandbox files, terminal output, structured artifacts, and the final provider-neutral Change Request.

Commercialize safely

Self-host the AGPL app today; use hosted xmode later for managed sandboxes, SSO, usage controls, and support.

Commercial hosted xmode

The open-source app is the product surface. Hosted xmode is the operational layer.

The core app remains open source. The hosted service can add managed sandboxes, GitHub App installation flow, private repository sync, SSO, audit retention, usage controls, and deployment support.

Managed sandboxes

Run work in maintained environments with logs, artifacts, browser/session support, and cleanup policy.

Enterprise integrations

GitHub/GitLab repository import, signed event intake, SSO, billing, and workspace administration.

Commercial support

Help with rollout, governance design, upgrades, action authoring, and production operations.

Latest writing

Field notes for governed agent work.

Long-form notes on structured runs, skill catalogs, sandboxes, and the operating model behind delegated software delivery.

All posts

Automation

Why software teams need governed agent pipelines

Agents are useful when they operate inside a system that captures intent, plans, approvals, logs, and reviewable changes.

Product

Skills, actions, and objectives in xmode

A skill describes reusable capability. An action turns that capability into a typed step that can run inside a pipeline.