The plugin includes a free, lightweight LLM model with no setup required. If you’ve never used AI in your workflow, it’s a great way to try it out. If you’ve been copy-pasting scripts into ChatGPT, this is a much more streamlined solution!
Note: Beta - may have bugs. Please use version control or a manual backup before starting.
We’ve added a ton of community requested features this release, made the AI smarter, and fixed a bunch of bugs. Here are some highlights:
New Features
TileMapLayer tools: New shape primitives for tilemaps: draw horizontal/vertical lines, stairs, and erase rectangles
Chat titles: Conversations now get AI-generated titles instead of just showing the first message
UI mode setting: Choose between main screen tab or side dock (now defaults to side dock)
New models: Kimi K2.5 Model and Claude Opus 4.6 support added
Auto Context Usage Widget: See your context usage right in the chat input to stay aware of token limits
Pixel Art Generator: Generate art with Retrodiffusion straight from inside Godot
Google OAuth: Sign in with your Google account for easier authentication
Flatpak Support: Ziva now works on Linux Flatpak installations!
Improvements
Settings dialog redesign: Tabs are now icons to fit better, unified upgrade/rate limit dialogs
Code blocks: Higher contrast for better readability
Input box: Now resizable
Chat header: Shows chat title instead of ID
Memory optimization: Reduced allocation churn on Linux builds
HiDPI Display Support: Fixed blurry rendering, scaling issues, and font rendering on high-resolution displays across all platforms (KDE, Gnome, etc.)
Installer
Performance: Reduced memory allocation churn in CEF texture updates on Linux
Physics Layers: Added physics layer support to configure_tileset_atlas tool
Bug Fixes
Linux release: Fixed missing CEF files
Chat messages: Fixed angle brackets being stripped from messages
I used to be in this camp as well. But when I set off to build an MMO on the side of my 9-5, I realized I simply didn’t have time for my ambition. Using AI is one way to bridge that gap.
@Dad3353 I respect your PoV, but there’s a difference between paying and managing a team and doing it yourself. AI doesn’t remove paying game devs from the equation, but it can empower each game dev to do more.
Updated to version 2.3.0 - Focused on stability and quality of life this time around.
Stability fixes (especially Windows):
Fixed copy/paste not working on Windows
Fixed installer extraction errors and ripgrep not found
Fixed save_scene crash during rapid parallel tool calls
Fixed get_scene_tree crash on deleted script references
Fixed edit_file silently succeeding without actually writing changes
Fixed dozens of silent fallback paths that hid errors instead of surfacing them
New stuff:
Undo button - Ziva now takes file snapshots before making changes, so you can revert with one click if something goes wrong
search_godot_docs tool - Ziva could previously look up class docs before answering, but now it has full access to all the docs, enabling it to be a much better teacher
run_tests tool - Ziva can now write and run tests for you
Message queuing - queue up messages while Ziva is still generating
Better model - You now can get GPT Mini (free), which is better than GPT Nano we were giving before
Also made some editor improvements:
“Files Modified on Disk” dialog auto-suppressed (no more reload popups every time a file changes)
Higher contrast syntax highlighting meeting WCAG AA standards
Hi! Is it possible to specify permissions for automatic actions for file types? For example, so that ‘read’ would be available for *.tscn, but not ‘edit_file’.
I mean, for now the agent often wants to use ‘edit_file’ for *.tscn, although I didn’t ask for it and I want changes only for scripts. That is, after each response, the agent sends a request with the ‘edit_file’ flag and I click prohibit for all scene files, every time.
It would be nice to be able to limit this more specifically. That is, to determine the scope space for the agent’s work.
Thanks for your answer!
P.S. There is a problem with the buffer, several times copying part of the response from the agent panel hung the program when trying to paste =]
The barrier to entry for making a game was never the time required to write code. Thats the easy part and if you cant do that fast then you need to work on that
All this is going to do is cause people to make bad games a little faster, as well as destroy their own skills in the process. What happens when this tool is no longer worked on/becomes more expensive?
Youre charging a subscription for a tool that uses AI flagship models, up to $200 a month.
Then you must buy an AI flagship model subscription on top of that. It feels scammy when pretty much all this is, is the equivalent of giving a bunch of user prompts to the AI (skills) on how to use Godot
If your problem with making an MMO was that you have a 9-5 and its going to take too long, then maybe dont make an MMO… Theres many games you can make, to free yourself from your 9-5, to focus on larger games
I also will say, there’s no way, even with pushing slop all day, that youll make an actually enjoyable MMO with this little human touch. It’s passionless, soulless, and eventually will result in 0 sales and a ton of wasted money. Which i assume is why you created this scam
If you dont enjoy the process of working on your art, and you try to pump it out as fast as possible, then you can expect other people to not enjoy it.
Its why there’s thousands upon thousands of games released every year on steam, yet only a few make it. Speed was NEVER the issue. In fact, the devs who went the slowest seem to have much better chances. Look at Dwarf Fortress, Stardew Valley, ETC compared to the average indie release. The difference is they spent the time required to make something enjoyable through and through
1/ Regarding Ziva shutting down one day: We’re actually going open source! I don’t have a timeline since we are mid-refactor to decouple the server, but we’re releasing open source so people are free to use it with their local LLMs and other AI subscriptions.
2/ You said “Then you must buy an AI flagship model subscription on top of that” - With Ziva, you do not need an AI subscription. Ziva has no dependencies, and (even when using the free tier), we provide access to all the AI models. The 200$/mo plan you’re referring to is for studios who want near unlimited usage of flagship models. That said, most people are fine on the 20$ tier. I’ve literally spent hundreds of hours optimizing things to make the 20$ tier worthwhile.
3/ I personally agree that coding is not always the bottleneck. But it is for some people. Ziva is not going to make a professional engineer suddenly have massive success on their game - but there are many indies and small studios using Ziva to build cleaner, more polished, and more ambitious projects than ever.
I also want to make something generally clear:
I’m are building Ziva as a response to Unity AI, and the upcoming AI for Unreal Engine. I love Godot, and want to give back to this community as much as I can, by making the best development tools possible. I understand AI is a hot topic. I’m not asking you to use it. But it should be available in the best form possible to people who want to use it.
In case it’s not clear by now, we are not trying to run a scam. I’m personally down several thousand USD since I started running Ziva, and I am happy to continue losing money as long as people enjoy it. Despite my bank account draining rapidly, I still take the time to donate 5% of all Ziva revenue to the Godot foundation.
Tl;dr - Ziva doesn’t require a 3rd party AI subscription, we are here to make awesome Godot tooling and help the ecosystem.
im not sure i understand how the economics work here, youre offering it for free, but also allowing “nearly unlimited” use of flagship models with a $200/month subscription? What is “nearly unlimited”? $200 is a very small amount, and if you’re not getting income from other places, how are you going to sustain this?
I also want to draw attention to how you only responded to 2 things I brought up, this isn’t a forward moving discussion anymore so Im not going to reply in the future