KodaCode 0.8.0 adds secure chat mode, dialog branching, and image support

KodaCode released version 0.8.0 and significantly reworked the interface behavior. The main focus is on more predictable UX: chat now follows agent logic, but works only with read-only tools, and multimodal models gained image upload capability directly in the dialog.

Safe Chat Mode The main change in the release is a new principle for chat mode operation.

Now it mirrors agent mode logic, but does not have access to tools that can change files or project state. This makes code base exploration significantly safer: you can view structure, read files, study dependencies and code behavior without the risk of accidentally running an edit in the wrong place. For teams where it's important to separate analysis from intervention, this mode reduces unnecessary tension and makes product behavior clearer from the first launch.

At the same time, Edit mode was removed from the product. KodaCode developers explain this simply: its scenarios are now fully covered by agent mode. In practice, this means fewer modes, less interface confusion, and less function duplication.

When a user has one main way to make changes and one safe way to explore the project, the barrier to entry becomes lower. This is not the most dramatic type of update, but these kinds of solutions usually have the strongest impact on daily work experience.

Branching and History The second important part of the release is dialog branching.

If the agent went off track, there's no longer a need to rewrite the request from scratch or scroll through a long session trying to recover a successful point. Now you can return to any tool call in the history, continue work from there, and create a new branch without affecting the original scenario. This is especially useful in tasks where you need to check several approaches in sequence: one for analysis, another for fixing, a third for an alternative hypothesis.

For developers and teams, this provides several practical benefits: easier rollback to a successful step without loss of context; more convenient comparison of alternative agent actions; easier debugging and repeating successful scenarios; safer experimentation with long chains of tools. Additionally, sessions can now be exported to JSON. This looks like a minor utility function, but it has many practical use cases: storing history, analyzing failed runs, sharing cases with colleagues, preparing reproducible examples for support or internal knowledge base.

In other words, KodaCode makes interaction with an agent not only interactive but also more portable. When results can be saved in a structured format, they're easier to work with outside of a single chat window.

Images in Chat A separate update concerns multimodal models.

For gemini-3-flash-preview, gpt-5.3-codex, kimi-k2.5, and gpt-5.

4, the chat gained the ability to attach images. This shifts some tasks from the "describe the picture in words" format to a more natural scenario: the user simply shows a screenshot, diagram, or mockup and immediately gets the model's response in the right context. For development tools, this is especially useful because many problems are explained faster visually than through long text.

The new feature opens several obvious scenarios: parsing screenshots of errors and UI bugs; analyzing layouts and interface states; reading diagrams, schemes, and documentation fragments; discussing model output results by image directly in one session. It's also important that support was added not abstractly "for everyone," but for a specific set of models. This shows a more careful approach to the product: multimodality is enabled where it's truly supported, rather than masking limitations with a unified interface.

For users, this is more useful than vague promises, because it's immediately clear which models can be used for visual scenarios without surprises.

What

This Means KodaCode 0.8.0 release is not about dramatic demonstrations but about control over the workflow. Safe chat mode, session branching, and multimodal attachments reduce the number of random errors and make agent work more predictable. For an AI development tool, this matters more than extra "wow" factor: users need not only a smart agent but also a clear environment where they can calmly explore, experiment, and step back.