FinchStory Updates

The most important change in this release is support for GPT6 Luna and GPT6 Sol. For users working with OpenAI / Codex inside FinchStory, that means the new models can be selected directly without manually maintaining a separate underlying environment.

When adding an OpenAI account in FinchStory, the confusing part is often not entering the device code, but knowing whether FinchStory is still connecting, has already failed, or needs you to act again.

Once account avatars started appearing in history, a more important requirement followed: messages from you, from different roles, and from role-to-role collaboration all need to keep the correct sender identity.

A brief network or service hiccup does not always need to end the whole request. This release improves recovery specifically for direct DeepSeek connections.

These releases turn the account avatar into something you can actually personalize: upload an image, crop it, preview it, and confirm before anything changes.

This release fixes a request-path issue that could make tool capability disappear before a request was actually sent, and unifies capability information across the same turn.

The main change in these releases is clearer FinchStory Loop permissions: whether a role can write files, reach outside the workspace, or run commands without approval is now much easier to understand. A few everyday controls were cleaned up at the same time.

Theme no longer means appearance mode. These releases add Classic and Premium, and either theme can be used with Light or Dark.

These releases refine the role creator and local-model equipment editing so role information stays stable and Developer Mode adjustments take fewer steps.

These releases reorganize role creation and OpenAI account information: the model list is complete again, new roles start at a safer speed, and account choice and subscription timing are easier to read.

When several OpenAI accounts are signed in at the same time, the important part is not simply whether they can all be added. They should stay independent, without a new login, re-login, or credential problem changing how another account is already being used.

These releases complete the main Seedance video workflow: choosing how to generate, tracking progress, stopping a job, handling failure, and saving and playing the result.

Equipment is no longer just extra text attached to a role. It can now be saved, ordered, bound, and checked for compatibility as its own configuration.

This release brings Ministral 3B into FinchStory's local-model path with text, image understanding, and native tool use.

This release moves the built-in editor from simply opening files toward being practical for real editing, especially with long files and repeated changes.

FinchStory needs to install a few required components during initial setup. The confusing part was not the installation itself, but knowing what stage it was in and whether it was safe to keep interacting.

This release makes one focused change: when FinchStory enters fullscreen, it removes unnecessary empty space above the left sidebar.

This release moves Windows support from simply launching toward working normally, filling in platform differences around runtimes, paths, file operations, and process cleanup.

This release group closes out several issues that directly interrupt workflow: restoring history features, preventing temporary outages from being mistaken for sign-out, and reducing jumps during session switches and attachment loading.

This release group focuses on restoring long sessions and reducing the work required when opening them. Expanded process history should not disappear after switching sessions, and a very long session should not need to load its entire history up front.

This release group continues refining the interface while making process display and interaction behavior more consistent across working styles. The goal is simple: changing styles should not mean learning a different set of interface rules.

This release group is dedicated to image preview: multi-image navigation, keyboard controls, wheel behavior, and more consistent sizing and centering.

This release group revisits nearly the entire sign-in and first-connection flow. The goal is simple: verification codes, OAuth return, sign-out, and getting back into the app should all be clearer.

This release group focuses on first-run setup, Runtime installation, and consistency across execution modes so new users have less to guess and everyday sessions behave more predictably.

This release group focuses on Activity, reasoning, and streaming behavior in longer sessions so progress feels more continuous and the interface does less unnecessary work.

Choosing the model is only part of the decision. How the task is carried out can change the experience just as much.

FinchStory has reached a new point in its development: applications are now officially open.

As features and model support change faster, remembering your current version and hunting down release information becomes unnecessary friction.

Once several accounts can stay signed in, the next question is how to keep account management from becoming work of its own.

Model providers move quickly, but the desktop app version should not decide when you can see a newly available model.

As you add more roles, putting all of their work through one Codex account can make quota the next bottleneck.

AI changes quickly, and the client should not be the last place to catch up.

Real work does not always happen on a perfect connection.

Real projects do not finish in one conversation.

As the number of AI options grows, model names alone become less useful for making a decision.

If local AI can only answer questions, its usefulness is still limited.

Not every piece of work needs to go to the cloud.

The hard part of meeting a new AI workspace is often not opening it—it is knowing where to begin.

Installing a Mac app should not require learning about signing, notarization, package formats, or system security first.

The same AI does not have to work in only one way.

As you use more AI, the friction shifts: more accounts, more connection methods, and more services that each feel like a separate tool.

The longer a conversation gets, the more messages and activity it has—but reopening it should not get slower just because the history is valuable. In 0.1.163, a real long conversation became ready to use in about 0.3 seconds instead of about 8 seconds.

After the previous release improved sign-in protection, Pi + ChatGPT could stop shortly after a conversation began. 0.1.162 restores the connection without giving Pi access to more account information.

FinchStory can connect to several AI services, but that should not mean every part of the app can see your login information. 0.1.161 limits where that information can go and who can receive it.

Picking a model or switching accounts is a small choice. It should not make the whole setup window grow, shrink, and jump around. 0.1.160 keeps those changes contained so you can stay focused on the role you are creating.

A handoff is more than changing one name. The activity title, current role, and member list all need to agree on who took over. 0.1.159 brings those states back into sync.

Projects often need a handoff.

When you hand off the next task, a role name alone is not enough. You need to know whether that role can still work, how much capacity remains, and whether its reasoning level and permissions fit what you are about to ask it to do.

Not every AI is good at the same kind of work.

When you add another way to run an AI agent, you should get more choice—not another installation, another sign-in, or another credential to look after.

As your project grows, the conversation should not get heavier every time you ask for the next step. You still need earlier work when you want it, without forcing every new reply to carry the whole history.

Editing a project file in place is useful only if you can trust what happens when you save. You should not lose newer work, end up with a half-written file, or forget where imported reference material came from.

When multiple AI agents work on the same project, the challenge is not adding more models. It is keeping roles clear, handoffs traceable, files controlled, and project context intact as the work moves forward.

A file entering AI context should not be an invisible action.

Often the problem is already visible. The friction is finding where to act next.

AI producing an answer does not mean the work is finished.