Changelog.

Security audit settings fit on one line, and nothing is sent for you

Choosing Security audit writes the start message into the composer with your settings, and the audit starts only when you send it.

The audit depth, check-ins, agent limit and models now sit on one line above the composer, each with a short explanation. The agent limit offers clear choices instead of a number box.

Each role can use its own model and thinking level, picked with the same model menu as the chat; roles you leave alone follow the chat's model. The start message waits in the composer until you send it, and appears once in the chat.

Agents can use a real browser

Agents can browse the web in a real browser in any chat, and you can watch, take over and replay what they did.

Install Browser from the Plugins page. The agent's browser runs on Cloudflare Browser Run, or on your workspace's machine while it is running, where it can open your own development servers.

It asks before it submits a form, uploads a file or opens a site the chat has not visited, and it can hand you the page for a sign-in or a captcha. The Browser tab beside the chat shows the live page, which you can take over, and a replay of every step.

Your plan includes browser minutes; past them, browser time is charged from your credit.

Security audits run through, and machine starts show in the chat

Security audits no longer stop at the coverage plan or the sandbox, and a chat now shows when its Linux machine is starting.

Security audits no longer stop before they begin hunting: the coverage plan saves, and the private sandbox starts on every machine. Audit agents are named in the chat by their role and number, and the audit settings show only the depth until you open Options, which holds check-ins, the agent limit and the models, laid out for phones too.

When a command needs the workspace's Linux machine and it is not running yet, the chat now says that the machine is starting and how long it has taken. A start usually takes seconds; on a server that has not run a machine yet it can take up to two minutes. Several commands that need the machine at once now start it once.

If the included models are briefly unavailable on our side, the chat now says so and that your allowance is unaffected, instead of saying your quota is used up.

If the studio restarts during an audit, ask the audit chat to continue: it picks up from its saved coverage, and work that was cut off is planned again.

If Security audit is installed in your studio, open Plugins and choose Update on it.

Check a security audit's open findings in the browser yourself

Findings the audit could not confirm now come with a supervised browser check, and you can record what you found in the audit.

Some findings only show in a real browser, such as cookie flags or a redirect after sign-in, and the audit's sandbox has none. The Audit sheet now lists these findings with what to check. Check in a browser opens a new chat with a checklist for the Browser plugin; you fill in your own deployment, such as staging, and send it when you are ready. Without the plugin, Install Browser takes you to it.

When you are done, choose Reproduced, Not reproduced or Inconclusive, and send the message it writes in the audit chat. After you approve it, your result is kept with the run's report, without changing the audit's findings. The audit's own agents no longer see tools from other plugins.

Chats keep working after a workspace restarts

A command that used more than a workspace's limits no longer leaves new chats failing.

When a command in the serverless shell ran past a workspace's limits, such as git status across a very large repository, the workspace restarted, and new chats in it failed until the account was restarted. They now reconnect on their own.

The studio has its own address

Your workspaces now open at studio.delorun.com, with a sign-in of their own, apart from your account and billing.

The studio, where your workspaces and chats live, now runs at its own address. It signs you in through delorun.com and keeps its own session, so code that runs in the studio, such as an installed plugin or an agent's command, can no longer reach your account, billing or sign-in methods.

Links to the old address take you to the new one, and you sign in once.

Connecting GitHub works when the app is already installed

If the Delo Run GitHub app was already installed on your account or organisation, connecting now finishes instead of stalling.

Connecting GitHub now finishes when the Delo Run app is already installed where your repositories are. The repository picker in setup also offers to authorise GitHub directly, so you can choose a repository without leaving it.

A changelog

Every change you can notice is now listed here, with an RSS feed.

This page lists what is new, what works better and what was fixed, newest first. Follow it in a feed reader at /changelog.xml.

Analytics, and your choice about it

Delo Run now measures how the site and the studio are used, so problems are found sooner, and you can turn it off.

Delo Run now uses PostHog, hosted in the EU, to see which pages and features are used and where errors happen. In the UK, the EEA and Switzerland, visits are counted without cookies. The studio is never recorded as a replay, and model calls are counted without their prompts or answers.

You can turn analytics off at any time on the privacy notice.