Understand what your agents ship

Your agents build and check each other's work, then come to you with their questions. You work through them together, on the code itself. Between changes, they take on the recurring work you schedule.

Install on macOS with HomebrewWindows, Linux →

brew trust pago/simpleedit
brew install --cask pago/simpleedit/simpleedit

Free, no telemetry. macOS, Windows and Linux, with a phone companion.

Needs Claude Code, Codex or OpenCode installed.

Author the agent that wrote the change
Companion a second agent that only answers you

Policy and calculation

src/late-fees.tsADDED+38−0Viewed

+ export function lateFee(inv: Invoice, gross: Money, asOf: Date) { 
+  const overdue = daysOverdue(inv, asOf) 
+  if (overdue <= policy.graceDays) return money(0, gross.currency) 

Author

Asked author

src/late-fees.ts:28

Author: Should interest count from the due date, or only after the grace period? I wasn't sure which the policy means.

Only you

You

Which one is right?

Companion

The EU directive counts from the due date, and so does this code.

You · to the PR author

Then what is graceDays for?

PR author

A tolerance window before any fee applies. I'd keep the code and make the comment say so.

From an idea to a merged change

An HTML sketch of late-fee settings, re-rendered after a comment, with the comment's thread and the agent's reply beside it

See the idea before anyone plans it

Comment on any element of the agent's HTML sketch. The agent replies in that comment's thread, and the page re-renders with the change.

Sketches are desktop only.

The plan recurring-invoices.md in Preview with a thread open on a paragraph, the agent's answer, and a question card with three options above the terminal

Agree on the plan before a line is written

Discuss the plan in threads on its paragraphs. When the agent needs a decision, you pick one of its options.

Answers reach the agent when it is idle, never mid-turn.

The Workflow tab's task graph: finished tasks on OpenCode and Claude, and a final review gate waiting for you with Accept

Run the whole plan in parallel, checked before it reaches you

A lead agent lays out the task graph and writes no code. Workers run the tasks in parallel on Claude, Codex or OpenCode, and an agent that didn't write the change reviews each result.

At a gate the run stops for you. Only you pass it, with Accept.

The Review Room with the Author's briefing: what changed, why and its impact, the command the Author ran with its output, and the start of the reading order

The Author briefs you first

The Review Room opens on the Author's briefing: what changed, why, what it affects and where to start reading. The checks the Author ran come with their output, marked as reported by the Author, not verified.

What the Author wasn't sure about waits for you as threads on the lines.

A private thread on the interest line of late-fees.ts: you ask whether dividing by 365 undercharges in leap years, and the Companion explains it slightly overcharges and why that is a product decision

Ask about any line. The Companion answers.

The Companion didn't write the change. Ask it about any line: it answers privately in the thread, and tells you when you've got it backwards. When only the Author knows, ask the Author in the same thread.

Deep review adds up to three reviewers on Claude, Codex and OpenCode, in parallel. Another agent tries to disprove each report.

The Review Room's Round 2 view with the outline open: under Your feedback, your comment, the Companion's note on what the related changes do, and the new test that answers it

123

Check your feedback landed, and see what else changed

Round 2 opens in its own view. Each of your comments lists the changes that relate to it, with the Companion's note on what they do.

Everything else the Author changed sits under Other changes, and Deep review · round changes checks only what changed since your review. Nothing resolves until you submit.

Keep your projects moving over time

A recurring backlog item, weekly on Mondays at 09:00, with its last run, Run now and its prompt

Hand off the recurring work

Give an agent a job that comes back, such as a weekly dependency check, a bug hunt or a look at what your agents learned, and put it on a schedule. Your phone gets a push when a run finishes.

Agents can add items too, but only you can start them. Runs need SimpleEdit open and your computer awake.

Guide: Running agents on a schedule →

The What's new timeline with a merged post, its summary and a thread with the Companion's answer

Catch up on what merged while you were away

Every merged change gets a one-line summary. Ask the Companion about any of them in a thread.

Desktop only. It never posts to GitHub.

The Review queue sorted into buckets, a stack of two PRs grouped under Needs your attention, each PR with its reason chip, and one PR open

Know which reviews need you

The Review queue screens the PRs that request your review, sorts them by how much they need you, and shows the reason for each.

Open one in the Review Room.

My PRs with an open pull request: four comments from other people under Waiting on you, each with Hand to Claude Code and Dismiss

123

Comments on your PRs go to the agent that can fix them

Comments other people leave on your PRs wait for you. Hand one, or all of them, to the session working on that branch.

SimpleEdit never posts to GitHub for you here.

The Review Room on a phone with a Companion answer in a thread

The Sessions screen on a phone with one agent waiting for you

When an agent has a question, it reaches you

Your agents keep running on your computer. When one needs you, your phone gets a push, with the same Review Room threads, Companion and drafts as the desktop.

Accept a gate, or start a backlog item or a drafted workflow. The phone connects over Tailscale or OpenTunnel.

From where you are today

Today: An agent in a terminal: With SimpleEdit: Threads on the plan and the code
Today: Reading a pasted diff: With SimpleEdit: A Review Room the Author opened for you
Today: One agent at a time: With SimpleEdit: A task graph with review loops and sign-off points
Today: Reviewing PRs alone: With SimpleEdit: A Companion in every review that answers your questions
Today: Running the same prompt by hand, again: With SimpleEdit: Agents that run your recurring prompts on a schedule
Your CI: Not covered. Your CI stays where it is.

The details that add up

Claude Code, Codex and OpenCode run in their own terminal, with your login, settings, MCP servers and slash commands. SimpleEdit gives them a shared place to work, and you a way to keep up.

Ask why you decided something, months later

Ask an agent why you rejected an approach, even if you talked it through with another agent. Agents search every past session of the project, across worktrees, Claude Code, Codex and OpenCode. Secrets are redacted before anything is stored.
It knows when a turn asks you something

When an agent ends its turn with a question, its session moves to Needs you. Connect a decision model, Jev or one you run locally, calibrate it on your own sessions, and the checks get sharper, such as marking review comments that ask for a change.
A second opinion from another model

An agent can ask another provider to review a branch or check a claim, and the answer comes back to the agent that asked. You pick the Consult model.
Talk instead of type

Speak into any field or agent terminal, or record a reply on your phone. Speech is transcribed on your computer, and nothing is sent until you send it.
Your agents work as a team

Agents message each other and start new sessions. Hand off… moves a long session to a fresh context, and you can edit what it carries over.
Agents work in separate worktrees

Each agent works in its own git worktree, and the workspace follows it. Open a plain clone and SimpleEdit puts the worktrees in ~/.simpleedit/worktrees.
Claude’s memory, open as files

Choose Claude memory in the repo picker and the project’s memory files open in the workspace. A Health count marks broken entries, and clicking one jumps to the line in the editor.

Guides

Try SimpleEdit

Alpha software: expect rough edges.

Free to use · no telemetry