For a while, it felt like every AI conversation was Anthropic this, Claude that. In our own work with Flashbag, OpenAI and Codex have made a strong comeback. GPT-6 Astra is the latest reason to give them another look.

But a month from now, Claude could release a model that changes the conversation again. We don't want you to rebuild your setup every time that happens.

01

Change the model. Keep what you've built.

Flashbag doesn't pick a permanent winner. Your agents, their memory, working context, and skills stay with you when you switch the model powering them. You can try something new without starting your team from scratch.

That's why this release focuses on Codex. It's where much of our work went this time, and we want the experience to feel right. When another model is the better fit for your work, you should be just as free to use it.

02

GPT-6 Astra is ready in the model picker

GPT-6 Astra is now available above GPT-5.6 Sol in Flashbag's model picker. It supports reasoning levels from Low through Ultra, plus Fast mode.

Choose it for an existing agent and put it to work with the context and skills you've already built. The video above is OpenAI's introduction to the model.

03

Check for updates from the Dashboard

On a Mac, checking for a Flashbag update used to mean heading to the app's menu in the macOS menu bar. Now there's another place to do it: the Dashboard.

This also works through Remote access, so you can check whether an update is available while using Flashbag remotely. A small change that makes keeping up with new releases easier.

04

A steadier experience, especially with Codex

Most of the work in 0.2.2 is under the hood. These fixes help conversations stay readable, recover cleanly, and keep work attached to the right place.

  • Resumed chats stay in sync. Live progress and tool activity remain on the correct message when a Codex chat resumes.
  • Remote chats recover properly. Reloaded chats show when work is still running, recover finished replies automatically, and tell you when no reply was saved.
  • Image generation keeps the conversation going. Codex image generation no longer causes chat turns to fail.
  • Flashbag sessions stay separate. They no longer appear in ChatGPT or ordinary Codex CLI history. Existing Codex users may need to sign in once through Flashbag.
  • Gordon handles Profiles more reliably. His tools use the Profile captured for each request. He also accepts uploads from every selected Profile while keeping his own files in the Default Profile.
  • A few everyday snags are gone. Voice dictation works in Linux AppImage installs, and selecting Agents works again after removing the Selected Profile. Ask Gordon is also available on failed turns to help diagnose problems.

For the full release notes, visit the changelog.

Flashbag 0.2.2 is available now.
Check for updates from your Dashboard, try Astra, and keep building with the agents you already know.