Agents can write code.IAS keeps humans in control.
IAS turns intent into governed action. This journey shows the engineering proof: the context, the checks, and the evidence behind every change.
reviewable
Every change stays reviewable.
Agent output resolves into one inspectable diff: real hunks, real file paths, nothing summarized away.
local execution
Your code stays in the repo.
IAS selects an approved local adapter for the run. The current default, codex-direct, runs Codex on your checkout; the Hub coordinates and records the run without holding your source.
context
Project memory travels with the run.
A context pack carries your standards, rules, and domain knowledge into the run. You see what is included and what is left out before anything executes.
approval
Nothing ships until a human says so.
Approval boundaries are part of the run, not a bolt-on. The gate shows the files, the checks, and the evidence, then waits for a person.
approval required
Apply change to main?
evidence
Checks stay attached to every change.
The run records the change as a commit, with its checks and decisions beside it. The audit trail is durable, not a screenshot.
see it in action
This is IAS Hub.
The operator surface for governed runs: context, policy, approvals, and evidence in one place.
request access
Keep humans in control.
Put your next code change through a governed run: context declared, checks applied, evidence kept. Request access, or send this page to the person who runs your agents.