Skip to content

Give IAS a goal.IAS keeps humans in control.

IAS turns rough intent into a useful outcome you can inspect. Start a local CLI run around Codex or Claude Code.

reviewable

Every run stays reviewable.

Inspect a run's outcome, assumptions, and artifacts without digging through a transcript. Runs are read-only by default; explicit writing tasks use an isolated worktree.

src/runtime/execute.ts
async function execute(plan) {
−const result = await agent.run(plan)
+const run = await harness.begin(plan)
+await run.recordEvidence(result)
+return run.gate(result)
}

Example

local execution

Run IAS on your machine.

Run records live in local state outside your repository. Selected context and task data go to your signed-in provider.

isolated worktree · local
$ ias run "<goal>" --mode workspace-write
→ writing in an isolated worktree
M src/runtime/execute.ts
changes ready for your review

Example

context

Only the context that can change the answer.

A run should gather the smallest high-signal context and show what it included and left out before it acts. You can correct that choice without restarting the goal.

context pack · 4 included · 2 omitted
project conventions.md
execute.ts (current)
adr/001-harness.md
recent decisions
unrelated legacy/*
node_modules/*

Example

approval

Irreversible steps wait for a person.

A person can inspect and redirect the work. Hard floors block. Ordinary reversible uncertainty does not become a permission queue.

approval required

Apply change to main?

4 files · 12 checks passed · evidence ready

Approve

Example

example evidence

The record is the result, not a screenshot.

A useful run returns the outcome, the material decisions, the artifacts, and what remains uncertain. You should not need transcript archaeology to know what happened.

commit recorded decision logged checks passed evidence recorded

historical evidence

See the governed run.

A historical Hub workboard shows a run's progress and the questions that needed a person's answer. Inspect local CLI runs through their intent, context, checks, and evidence.

Historical IAS Hub workboard
IAS logo

next step

Read how a run works.

Read the docs to see how a local run handles intent, context, checks, and evidence. Share the tour with the person who runs your agents.

Read the docs