AgentSDEDocs
Open console

How AgentSDE works

AgentSDE runs AI agents inside your own repository. An issue comes in, moves through a pipeline of phases, and an agent opens a pull request on your branch, under your reviewers and your checks.

6 min readUpdated 2026-08-04 · Priya NEdit this page

#The loop

Everything AgentSDE does is one loop, repeated. You bring the issue and the repository; the agent brings the work; your existing review process decides what ships.

1

An issue arrives

Work starts from a tracker you already use · GitHub Issues, Jira, or Bitbucket. An intake trigger decides which issues an agent picks up, so nothing runs until you say so.

2

An agent takes it through phases

A run moves through an ordered set of phases · restate the brief, plan the change, deliver a diff, review it, and compound what was learned. Each phase leaves an artefact you can read.

3

A pull request opens

The agent pushes to its own branch and opens a pull request, requesting the reviewers your branch protection already requires. You review it like any other contributor’s.

Nothing bypasses your controls

Agents are contributors, not admins. Branch protection, required reviewers and required checks apply to their pull requests exactly as they do to a human’s. There is no path to a protected branch that skips review.

#What you configure

Most teams start on the defaults and change one thing at a time.

SurfaceWhat it decides
Intake triggersWhich issues start a run, and how
Phase designThe phases a run moves through, and what each one does
PersonaHow the agent writes and what it will and will not do
EnvironmentsWhere a run executes and what it may touch

#Where to go next