Amp brings Portals to local machines: see your dev server right in the thread
Portals in Amp used to work only in cloud orbs. Now they are available on your local machines too: you ask the agent to start a dev server or build a site, and the result opens in the thread's Portal tab, wherever you are. In the announcement's example, an agent on a Mac runner built a disk usage breakdown page: 241 GB used out of 383 GB, with 155 GB taken up by simulators and Xcode data.
AI-processed from Amp News (Sourcegraph); edited by Hamidun News
A familiar situation: the agent wrote a frontend and started a dev server, and now you have to open a terminal, hunt for the port and check what came out. If the agent is running on another machine, say a Mac at home while you sit in a cafe with a laptop, the task turns into a small quest. The Amp team at Sourcegraph is closing that gap: Portals, which used to live only in cloud orbs, now work on your own runner as well.
What changed
A Portal is a tab next to the thread that opens a web page created by the agent. Previously it was available when the agent worked in an orb. Now the same mechanism works on local machines: you ask Amp to start a dev server or build a site, and you open the result in the Portal tab in the thread. An important detail from the announcement is "no matter where you are." In other words, the runner can sit on one machine while you view the result from another.
Example from the announcement
The Amp developers showed a simple but telling case. In a thread on a macOS runner, the user asks for an overview page on Mac file system usage and to show it in a portal. The agent replies with a link to a Disk Usage portal, and a small web app opens alongside: 241 GB used out of 383 GB, a capacity bar split by category, and a donut chart showing that simulators and Xcode data take up 155 GB.
The value is not in the chart itself but in the loop: a question, code, a live page, and you immediately look at the result and refine the task. One-off internal dashboards that nobody would have built by hand before become almost free. Need to figure out what ate the disk, look at build metrics or check the layout? Ask the agent and you get a page, not a wall of text in chat.
How to enable it and who will see it
There are two ways. In the macOS app, open Settings › Runner and turn on Allow Portals. If you work from the terminal, start the runner with the command amp --no-tui --remote-control-portals.
By default, your runner's portals are available only to you. To show them to colleagues, start a shared runner with the --share flag or turn on Share With Workspace in the macOS app. Then any workspace member can open that runner's portals. Details are in the Amp documentation under Portals on Your Runner.
Limitations and practical takeaways
Some things to keep in mind. The option is off by default, and that is the right call: you decide whether to let the agent expose web pages from your machine. When you enable sharing for the whole workspace, remember that colleagues will see everything the agent shows in the portal. And on a local machine, anything can end up on the page: paths, project names, data from the file system. The disk report from the example is harmless on a personal Mac, but it should not be shown to an entire workspace without checking. The announcement gives no details on how access works or what technical limits portals have, so consult the documentation for the nuances.
The practical takeaway is simple. If you already use Amp with a local runner, turn on Allow Portals and try it on a task where you previously had to start a server by hand: a frontend preview, a dependency report, a log dashboard. If you work in a team, start with personal mode and enable --share only for runners that hold nothing sensitive.
And if you build agents yourself, take note of the idea. The ability to show a result as a live page rather than text in chat is a separate and badly underrated feature. It shortens the path from request to checking the result with your own eyes, and that is exactly the step where time usually gets lost.
- Nov 12, 2026Claude in Chrome requires approval for every click on category 3 sites
- Nov 12, 2026Norwest raises record $4 billion fund, betting on agents, voice and physical AI
- Nov 12, 2026AWS shows how 100% retrieval precision can hide an AI agent's failure
- Nov 12, 2026Garage Eight dropped peer approvals for 69 MRs in an internal service using AI review
Key Questions Answered
What changed?
A Portal is a tab next to the thread that opens a web page created by the agent. Previously it was available when the agent worked in an orb. Now the same mechanism works on local machines: you ask Amp to start a dev server or build a site, and you open the result in the Portal tab in the thread. An important detail from the announcement is "no matter where you are." In other words, the runner can sit on one machine while you view the result from another.
What are the key details on Example from the announcement?
The Amp developers showed a simple but telling case. In a thread on a macOS runner, the user asks for an overview page on Mac file system usage and to show it in a portal. The agent replies with a link to a Disk Usage portal, and a small web app opens alongside: 241 GB used out of 383 GB, a capacity bar split by category, and a donut chart showing that simulators and Xcode data take up 155 GB.
How to enable it and who will see it?
There are two ways. In the macOS app, open Settings › Runner and turn on Allow Portals. If you work from the terminal, start the runner with the command amp --no-tui --remote-control-portals. By default, your runner's portals are available only to you. To show them to colleagues, start a shared runner with the --share flag or turn on Share With Workspace in the macOS app. Then any workspace member can open that runner's portals. Details are in the Amp documentation under Portals on Your Runner.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.