When an agent finishes a job, you want to read what it made. Flashbag 0.3.0 gives that answer more room, keeps the steps within reach, and adds a new way to choose what powers your agents.
Released on September 17, this update brings a redesigned chat, experimental OpenRouter access through Pi, and better tools for organizing Skills and managing defaults across your team. Here's what changes in everyday use.
The answer takes centre stage
Chat now opens with Minimal activity as the default. When work finishes, thinking, tool activity, and intermediate replies collapse, leaving the final response as the main thing to read.
The steps are still there. Open View Activity whenever you want to see how your agent got to its answer. Prefer to follow everything? Choose Full activity in Settings to keep those details visible.
Reply controls now sit together beneath each response. Copy the final answer, check elapsed time and token usage, inspect activity, or ask Gordon for help from the same place. You can also choose your own message colour and avatar in Settings.
Images get a proper gallery with thumbnails, keyboard and swipe navigation, and downloads for one image or the whole set. File attachments show clearer names, locations, and file types, so it's easier to find the output you need.
OpenRouter comes to Flashbag through Pi
Pi is a new, experimental way to run agents in Flashbag. It connects Agents and pipeline helpers to ChatGPT or OpenRouter, giving you another option alongside Claude Code and Codex.
The idea fits how we build Flashbag: you choose the model and the way it runs, while your agents and the setup you've built remain at the centre of your work.
To try it, open Settings → Experimental and enable Pi support. Then configure it under CLIs.
This is intended for advanced testing. Claude Code and Codex remain available by default.
You can also turn Claude Code and Codex on or off, or choose between installed copies, without restarting Flashbag. Pi offers the same controls once enabled.
Organize Skills around the team that uses them
As your collection grows, managing Skills one by one becomes slow. In 0.3.0, you can edit them individually or in bulk, organize them into Groups, and filter by Skill Group, Agent, or Agent Group.
In the walkthrough, I put daily emails and email sequences into a Copywriting Skills group. That gives you a focused view of related Skills instead of searching through the whole collection.
You can also assign Skills to an Agent Group. Those assignments follow Agents as they join or leave, so group membership determines which shared Skills they receive.
Skills created or edited outside Flashbag now refresh reliably too. Damaged settings no longer hide healthy Skills or accidentally grant access, and interrupted moves preserve access restrictions.
Set your defaults once. Adjust where it matters.
You can now set app-wide defaults for Preload, model, and reasoning. Agents using those defaults follow later changes automatically, so updating your usual setup doesn't mean editing every Agent.
For example, if you use both Claude and Codex and hit a subscription limit, you can change your preferred model in Settings. Agents using that default follow the change together. Agents with their own model choice keep it.
Preload controls how much context an Agent starts with. You can customize Preload and Pipeline independently for each Agent, keeping exceptions where you need them.
Change model and reasoning choices directly in chat, or Preload on the Agents page. Choices save immediately, and Reset restores your configured defaults.
Zoom controls also replace layout density, letting you choose the app size that feels right on each computer.
See what a Profile update will change
Checking for a Profile update now automatically prepares a review. Agents and Skills are grouped, folders are browsable, full file comparisons are available, and conflicts are highlighted before you approve.
You get a clearer view of what's arriving and where it overlaps with your own work. Update Flashbag before receiving Profile updates that include Agent Group Skill assignments or saved Skill Group names and order.
A smoother experience in longer, busier chats
This release also fixes several issues that showed up during everyday work:
- Image-heavy chats keep moving. Fixed hangs and excessive diagnostic event-log growth.
- Codex jobs stay independent. Stopping a reply, hitting an error in one conversation, or refreshing sign-in status no longer interrupts other running Codex work.
- Long chats use less memory. Switching conversations preserves your reading position and access to older messages. Loading image previews no longer shifts reopened chats.
- Startup stays responsive. You can keep using Flashbag while Remote access connects.
- Updates recover more cleanly. Claude Code, Codex, and Pi updates take effect without restarting. Refresh can recover interrupted updates, and canceled updates stop running in the background.
- Sign-in problems are clearer. Warnings appear across Flashbag, and Refresh clears them when access returns. Application updates also reject unsigned releases before downloading them.
There is also internal preparation for upcoming Windows and server support. Those are still ahead; this release lays groundwork for them.
For every change and compatibility note, visit the full changelog.
Flashbag 0.3.0 is available now.
Check for updates from the Dashboard and try the redesigned chat with your own agents.