Skip to content
Learn Kiro.

Concepts · Chapter 23 of 63

Kiro Web

Kiro Web went generally available on September 1, 2026. Plans, GitHub and GitLab support, Automations, autonomous mode, Memory, and config sync.

All levels 5 min read last reviewed 2026-09-04

◎ Learning objective

Decide when the browser surface is the right one, run an autonomous session or an automation, and know which plans and regions apply.

Kiro Web at app.kiro.dev runs the agent in your browser against repositories on GitHub or GitLab, with nothing installed locally. It reached general availability on September 1, 2026, and now runs on every paid plan rather than as a preview.

What is Kiro Web?

You connect a repository, describe a change, and a cloud agent implements it. The result arrives as a pull request or merge request you review like any other. There is no local checkout, no install, and no machine of yours doing the work.

Status, plainly. Generally available since September 1, 2026. Eligible plans are Pro, Pro+, Pro Max, and Power; the Free plan does not include it. Git providers are GitHub and GitLab. With AWS Identity Center sign-in, it runs in us-east-1 only. Credits are consumed the same regardless of surface, so Web spends from the same pool as your IDE and CLI.

Why it exists

Three audiences, one surface.

People with no local setup. A reviewer, a technical PM, someone on a borrowed laptop. The browser is the only place the work can happen at all.

Work that does not need your machine. Bumping a dependency across six repositories does not benefit from being watched in an editor. It benefits from turning into six pull requests while you do something else.

Recurring work. A weekly dependency sweep or a nightly flaky-test triage is a job for a schedule, not a person remembering. Automations exist for exactly that.

How it works

Ordinary sessions. Chat, the agent works, a PR appears. Sessions get agent-generated titles (August 27, 2026), and renaming one manually prevents the title from being overwritten. Since August 31, 2026 you can choose a repository branch when creating a session, which is the one thing plain cloud sessions still cannot do.

Autonomous mode. A toggle in the chat input, off by default. When on, the agent owns the outcome end to end and runs through phases: clarification, then planning, then execution by specialized sub-agents inside an isolated sandbox, ending in a pull request. The model is chosen automatically. A session can enter a needs attention state when it genuinely requires you, and you can steer from the pull request itself with the comment commands /kiro all and /kiro fix.

Automations. Scheduled autonomous runs, on Pro and above. Hourly, daily, or a CRON expression; up to five schedules per automation; times in UTC; one or more repositories; a prompt of up to 10,000 characters. Each run produces a pull request.

Memory. Kiro Web learns from the pull request feedback given by the person who created a task. That memory is per user and applies across repositories. Under Settings → Memory you can view and delete entries, though not edit them.

Review without leaving. The Reviews panel surfaces a session’s GitHub pull requests with status, commits, and diffs. Playwright and agent-browser support (July 2026) let the agent open the app it just built in a real browser, click through it, and verify the UI. There is a workspace file explorer for browsing what a session created, with syntax highlighting.

Configuration sync. Since September 1, 2026, cloud configuration defined in Web (agents, MCP servers, powers, steering, skills, hooks) syncs down to local IDE and CLI sessions. Cloud content stays in the cloud rather than overwriting local files: it appears locally as a read-only preview and stays editable in the browser. On the local side this needs IDE v1.0.437 or CLI v2.21.0.

Enterprise sign-in. Teams can authenticate through Okta or Microsoft Entra ID. Setup requires adding https://app.kiro.dev/signin/oauth to the OIDC application and enabling cloud sessions.

Common mistakes

Reviewing a cloud PR less carefully than a local diff. You did not watch this one being written, which is an argument for more scrutiny, not less. The fix: read the diff before the description.

Pointing an automation at a repository you do not trust. Kiro’s own documentation is direct about this: only select repositories you trust, because the agent follows instructions found in repository code. The fix: treat repository contents as input to the agent, and scope automations to code your team controls.

Assuming Free covers it. Web is a paid-plan feature. The fix: check the plan before planning a workshop around it. See Credits & Pricing.

Editing cloud configuration locally. Synced cloud items are read-only previews in the IDE and CLI on purpose. The fix: edit them in Web, where they live.

Permissions and safety

Two risks deserve naming. The first is prompt injection through repository content, which is why the trusted-repository warning exists: a file in a repository can contain text aimed at the agent. The second is unattended spend: an automation on an hourly schedule is a standing credit commitment with nobody watching each run.

The mitigations are ordinary. Scope automations narrowly and give each one a prompt that describes a bounded job. Watch credit usage after adding a schedule rather than at the end of the month. Admins can now export usage metrics to OpenTelemetry (September 1, 2026), which surfaces per-user adoption and credit consumption. And since Kiro is covered by the AWS ISO/IEC 27001:2022 certification as of the same date, the compliance question has a documented answer.

Where it fits

Web is the surface built directly on cloud sessions, and it adds the pieces cloud sessions lack: branch selection, Automations, autonomous mode, Memory, and PR review. The IDE and CLI are the local surfaces, and since September 2026 they read cloud configuration from Web. The iOS app steers cloud sessions from a phone and syncs with Web.

Kiro Crew is the one to keep separate in your head. Web is AWS-hosted, repository-shaped, and PR-oriented. Crew is an open-source agent that runs on hardware you own, persistent rather than session-shaped, and reaches chat platforms rather than pull requests.

Learn more

Frequently asked questions

Is Kiro Web still a preview?

No. Kiro Web reached general availability on September 1, 2026. It runs on the paid plans, Pro, Pro+, Pro Max, and Power, and is not available on the Free plan.

Does Kiro Web cost extra?

No separate charge is listed. Kiro's documentation states that credits are consumed the same regardless of surface, so Web draws from the same monthly credit pool as the IDE and the CLI. You do need a paid plan to use it at all.

Which git providers does Kiro Web support?

GitHub and GitLab. Sessions run against a repository you connect, and the result arrives as a pull request or merge request. With AWS Identity Center sign-in, Kiro Web runs in the us-east-1 region only.

What is autonomous mode in Kiro Web?

Autonomous mode is a toggle in the chat input, off by default, where the agent owns the task end to end. It runs a clarification phase, then planning, then execution by specialized sub-agents inside an isolated sandbox, and opens a pull request. The model is chosen automatically. A session can enter a needs-attention state when it requires you.

What are Kiro Web Automations?

Automations are scheduled autonomous runs, available on Pro and above. You can schedule hourly, daily, or with a CRON expression, up to five schedules per automation, in UTC, against one or more repositories, with a prompt of up to 10,000 characters. The output is a pull request.

☰ Chapter summary

  • Kiro Web at app.kiro.dev runs the agent against GitHub and GitLab repositories with nothing installed locally.
  • It reached general availability on September 1, 2026, on Pro, Pro+, Pro Max, and Power, and is not on the Free plan.
  • Autonomous mode runs clarification, planning, and sandboxed execution by sub-agents, and delivers a pull request.
  • Automations schedule autonomous runs hourly, daily, or on CRON, up to five schedules each, with the output as a PR.
  • Memory learns from the task creator's PR feedback and is viewable and deletable per user under Settings.
  • Cloud configuration sync pushes Web-defined agents, MCP servers, powers, steering, skills, and hooks down to local sessions.

All chapter summaries are collected on the revision page.

Was this chapter helpful?