# Diff review: read every AI change, send notes back, ship what you approve

> Diff review in Tallos is the loop for checking code that AI agents write. Every workspace has a diff against the branch it started from. You leave notes on any line, send them to the agent as one batch, and repeat until the diff is clean. Then you commit, push and open the pull request, with GitHub, GitLab, Linear and Jira linked to the same workspace.

- Canonical: https://runtallos.com/features/diff-review
- Português: https://runtallos.com/pt/recursos/revisao-de-diff.md
- Updated: 2026-09-29
- Section: Features

## Key facts

- A diff per workspace, against its start branch
- Line notes sent to an agent in one batch
- Commit, push and open a PR or MR in the app
- GitHub checks, reviews and comments inline
- Linear and Jira issues linked to the workspace

## What is diff review in Tallos?

**Diff review** is the set of tools Tallos gives you to read, correct and ship what an agent changed. Because every task runs in its own [parallel workspace](https://runtallos.com/features/parallel-workspaces), each workspace has its own diff: only that agent's changes, compared with the branch it started from.

The goal is to review like a lead, not a babysitter. You read the whole change once, write down what must change, hand it back, and only merge what you approve. The guide to [reviewing AI-generated code](https://runtallos.com/learn/review-ai-generated-code) covers what to look for in each hunk.

## What does the diff viewer show?

The diff viewer is built for full reviews, not quick glances. By default it compares the workspace with its start branch; from the toolbar you can compare against any commit, branch or the repo's base branch.

- **Combined diff** across staged, unstaged and untracked files, with a resizable file tree beside the hunks.
- **Inline or side-by-side** layouts, line numbers on both sides and optional word wrap.
- **HTML preview**: open a changed HTML file in a side browser split with **Open Preview to the Side**, using the [built-in browser](https://runtallos.com/features/built-in-browser).
- **Merge conflicts** in a three-way view, with **Resolve with AI** next to **Review conflicts** when you want an agent to take the first pass.
- **Keyboard navigation**: `F7` and `Shift+F7` jump to the next and previous change.

## How do I send review comments back to the agent?

Leave notes on the lines you want changed, then send them all to an agent at once. Batching keeps the feedback coherent: one round of thinking, one revision pass.

1. **Open the workspace diff** — Go through each changed file. For each hunk, ask whether the change is necessary, minimal and consistent with the rest of the file.
2. **Add a note on a line** — Hover a line and click the **+** in the gutter, or select a range. Write the feedback in full sentences; Markdown works.
3. **Send the notes** — Open **Send notes to** and pick the **Active agent session** in that workspace, or start a **New agent**. Tallos builds one prompt with every note anchored to its line.
4. **Let the agent revise** — Watch it work in the [chat or terminal](https://runtallos.com/features/native-chat). Notes stay pinned to their lines, so you can check each fix.
5. **Resolve and repeat** — Resolve the notes that are fixed and leave follow-ups on the rest. Unresolved notes go out with the next batch. When the diff is clean, commit.

> Download Tallos and review your next AI diff line by line. → https://runtallos.com/signup

## How do I commit, push and open a pull request?

The Source Control panel sits next to the diff, and its main button always shows the next useful step: **Stage All**, then **Commit**, then **Push**, **Pull**, **Sync** or **Create PR**.

| Step | What Tallos does |
| --- | --- |
| **Commit** | Stage files and write a message, or use **Generate with AI** to draft one from the staged changes. Your repo's pre-commit hooks run as usual. |
| **Fix a failed commit** | **Fix with AI** starts your default agent with the hook output, the attempted message and the staged files. It is asked to repair, not to bypass hooks. |
| **Push** | Pushes the branch and sets the upstream the first time. Never a silent force push; **Force Push** is a separate, labeled action that uses `--force-with-lease`. |
| **Open the PR or MR** | Confirm base branch, title, description and draft state, or generate the details with AI. GitHub, GitLab and Bitbucket Cloud are supported. |
| **Stack on GitHub** | When the base already has an open PR, choose **Stack this PR above #N**. |

_Shipping a workspace from the Source Control panel_

## Which review and task tools does Tallos connect to?

Tallos links each workspace to the pull request and the ticket it belongs to, so review state and task state sit next to the code.

| Surface | What you can do in Tallos |
| --- | --- |
| **GitHub** | Checks, reviews and comments in a PR tab; reply in review threads; **Fix broken checks** hands failed checks to an agent; auto-merge or merge queue; issues and GitHub Projects under Tasks. |
| **GitLab** | Merge requests and issues in the same review surface; pipeline jobs, including child pipelines, with their logs in the Checks panel. |
| **Linear** | Browse, create and update issues (status, assignee, priority, labels, estimate); create a workspace from an issue using Linear's branch name; images in the issue go into the agent's prompt. |
| **Jira** | Cloud or self-hosted; edit status, priority, assignee and custom fields; comment; create a workspace from an issue. Credentials are encrypted with the system keychain. |

_Review and task integrations in Tallos_

Bitbucket Cloud, Azure DevOps and Gitea pull requests also appear in the workspace sidebar and the Checks panel.

## How does diff review work with squads?

When a [squad](https://runtallos.com/features/squads) finishes, every member's work sits in its own worktree. The final review lists results by worktree with **View changes**, which opens the same diff viewer. Annotate, send notes to an agent, then use **Open to create PR** on the result you keep. Nothing is merged until you decide.

> **Why this matters** — Editors were built for one person typing. When several agents write code at once, the bottleneck moves to review. See [ADE vs IDE](https://runtallos.com/compare/ade-vs-ide) for how an agent development environment is shaped around that.

## Frequently asked questions

### How do I review code written by Claude Code or Codex before merging?

Open the workspace diff in Tallos, add notes on the lines you want changed, and send them to the agent in one batch. Repeat until the diff is clean, then commit, push and open the pull request.

### Can I send line comments back to an AI agent?

Yes. Add notes on any line or range of the diff, then use Send notes to and pick the active agent session or a new agent. Tallos sends every note in one prompt, anchored to its line.

### Does Tallos ever force-push my branch?

Not silently. Push never falls back to force; Force push with lease is a separate action that shows how many commits it replaces and uses `--force-with-lease`.

### Can I see GitHub checks and review comments inside Tallos?

Yes. Linked GitHub pull requests open in a PR tab with checks, reviews and comments, and failed checks can go straight to an agent with Fix broken checks.

### Does Tallos work with Linear and Jira?

Yes. Connect Linear with a personal API token and Jira Cloud or self-hosted with its credentials, then browse issues, update them and create a workspace from any issue.

### Can an agent write my commit message?

Yes. Generate with AI drafts a commit message from the staged changes, and the pull request dialog can generate the title and description. You review both before submitting.

---

**Run your agents in parallel with Tallos** — Claude Code, Codex, Gemini and 30+ agents — each in its own workspace, with squads, chat, terminals and review in one app. Uses the subscriptions you already have.

https://runtallos.com/signup (macOS 13+ · Windows 10+)