October 26Oct 26 Author @Jimbo.Automates , it's a known issue with all gemini models that I tested with. It times out because it claims that the servers are busy and we should try later. It might be that our end point is incorrect, that said, it'll always for the 2nd or 3rd time.With kind regards,Michel
October 26Oct 26 19 minutes ago, Michel Kohanim said:@Jimbo.Automates , it's a known issue with all gemini models that I tested with. It times out because it claims that the servers are busy and we should try later. It might be that our end point is incorrect, that said, it'll always for the 2nd or 3rd time.With kind regards,MichelThanks @Michel Kohanim but it's happened 4 times in a row, also seeing these errors2026-10-25 18:00:06.879 Thread-11 (main) udi_interface ERROR udi_interface:write: 2026-10-25 18:00:06 | WARNING | unified.adapters.gemini_adapter | Gemini streamGenerateContent failed, falling back to generateContent:2026-10-25 18:03:58.965 Thread-11 (main) udi_interface ERROR udi_interface:write: 2026-10-25 18:03:58 | WARNING | unified.adapters.gemini_adapter | Gemini streamGenerateContent failed, falling back to generateContent:2026-10-25 18:04:59.161 Thread-11 (main) udi_interface ERROR udi_interface:write: 2026-10-25 18:04:59 | ERROR | unified.run_unified_runtime | unified: turn failed:
October 26Oct 26 I had similar issues but seems you got further. What does your config look like for model etc
October 26Oct 26 Tried different models, both give 503:2026-10-25 18:22:55.860 Thread-11 (main) udi_interface DEBUG udi_interface:write: 2026-10-25 18:22:55 | INFO | httpx | HTTP Request: POST https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-pro-preview:streamGenerateContent?alt=sse&key=__REDACTED___ "HTTP/1.1 503 Service Unavailable" 2026-10-25 18:22:55.861 Thread-11 (main) udi_interface ERROR udi_interface:write: 2026-10-25 18:22:55 | WARNING | unified.adapters.gemini_adapter | Gemini streamGenerateContent failed, falling back to generateContent: Server error '503 Service Unavailable' for url 'https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-pro-preview:streamGenerateContent?alt=sse&key=AQ.Ab8RN6L63oYj-tJ0134CTo_VuK1su2gu1j6upwK6wk5MEnlIQw' For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/503 Response body: { "error": { "code": 503, "message": "This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.", "status": "UNAVAILABLE" } } Gemini said the url should be as below, generateContent instead of streamGenerateContent https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-pro-preview:generateContent?key=YOUR_KEY
October 26Oct 26 2 hours ago, photogeek54 said:I have the eisy-ai plugin configured with the default json and an anthropic API key. The plugin is started and the log appears to show it is connected and running.I can't find the interactive chat to start using the plugin. I don't see anything in the eisy-ui nodes, nothing under Eisy-ui plugins, nothing in the eisy-UI menu.What really helps me move forward are examples! some documentation showing a simple WORKING example of using eisy-ai for a few queries would really help. Example building a simple routine, example querying what the heating systems has been doing the last few days, example of what you would ask for to build a simple plugin. What kids of wording would you use to do these things?Examples are so helpful!ThanksThere is a star like icon in upper right corner you press. That open interaction window
October 26Oct 26 1 minute ago, Panda88 said:I had similar issues but seems you got further. What does your config look like for model etcI have a Pro subscription, and Gemini said I can use these:With a Google AI Pro subscription, your API key unlocks the paid tier in Google AI Studio, granting you access to Google's flagship "Pro" tier models that are unavailable to free users. It also removes the lower rate limits found on the free tier and ensures your API prompts are not used to train Google's models.You have access to the following primary models via the API:Pro Models (Advanced Reasoning)Gemini 3.1 Pro Preview (gemini-3.1-pro-preview): The latest flagship model with a 2-million token context window, designed for complex reasoning and coding.Gemini 2.5 Pro (gemini-2.5-pro): The previous generation advanced model.Flash Models (Fast & Multimodal)Gemini 3.8 Flash (gemini-3.8-flash): The newest, most intelligent Flash model engineered for long-horizon software engineering and enterprise workflows.Gemini 3.6 Flash and 3.7 Flash (gemini-3.6-flash, gemini-3.7-flash): Previous generation models for rapid multimodal tasks.Gemini 3.5 Flash and 2.5 Flash (gemini-3.5-flash, gemini-2.5-flash).Flash-Lite Models (High-Volume & Cost-Efficient)Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite): The fastest, most cost-effective 3.5 model for high-throughput execution.Gemini 3.1 Flash-Lite and 2.5 Flash-Lite (gemini-3.1-flash-lite, gemini-2.5-flash-lite).
October 26Oct 26 @Michel Kohanim Gemini says:The fact that you are now seeing this same 503 Service Unavailable error across multiple models—including the high-capacity gemini-3.8-flash—confirms that this is a broader, temporary platform-wide capacity issue on Google's end, not an issue with your Polyglot v3 (PG3x) plugin or API key.Here is what is happening and how to handle it in your adapter:1. It is a Server-Side Capacity IssueA 503 error specifically means that Google's servers simply do not have enough computing power available at this exact moment to service the request. It is distinct from a 429 error, which would mean you hit your personal rate limit. Because you are getting 503s on both Pro and Flash models, the Generative AI API is likely experiencing a significant usage spike or temporary degradation. Upgrading to a paid tier does not resolve 503 errors, as they affect all users when server capacity is exhausted.2. Implement Exponential Backoff Since your gemini_adapter.py is already smart enough to catch the streaming failure and fall back to standard generateContent, the next step for a robust node server is to implement a retry loop with exponential backoff.Instead of failing the turn immediately when generateContent returns a 503, your code should:Wait 2 seconds, then retry.If it fails again, wait 4 seconds.If it fails again, wait 8 seconds.Add a bit of "random jitter" (e.g., waiting 2.3 seconds instead of exactly 2.0) to prevent your retries from perfectly syncing up with other users retrying at the same time.3. Wait it OutBecause this is entirely on Google's infrastructure, if exponential backoff still exhausts your retry limit, your only recourse is to let the plugin fail gracefully and wait for the demand spike to subside. These spikes usually resolve as load balancing adjusts to the traffic.
October 26Oct 26 Found one that works: gemini-3.5-flash-liteBut would still prefer to use the more advanced models, but will play with this one.
October 26Oct 26 I have pro as well, but Gemini told me the API is separate from the Pro subscription - DId you manage to get a key with your pro-subscription?
October 26Oct 26 Author @Jimbo.Automates ,The keyword is "AI Studio" ... this is the same issue with Claude an OpenAI and Grok: you cannot use your subscriptions for API usage. You need a different plan.We are working on an MCP server. This way, you can use your subscriptions, but it's proven even more difficult because all the agentic loop is handled by the MCP host (i.e. Claude Desktop, etc.) + it has a max number of tokens which are way too low for larger installations.We also sent an email to Anthropic to see whether or not we can have another subscription based contract with them. Haven't heard back, and I don't think they will: we are too small for them.And, finally, we have a new release coming with these features:Fileviewer that gets updated based on the files that are changedMore optimized use of tokens using cache (for LLMs that support cache)Plugin directory has changed to /tmp/eisy-ai ... so, for those who already have plugins, you will need to copy (manually) to /tmp/eisy-ai:cp -r /home/admin/eisy-ai /tmpsudo chown -R admin:polyglot /tmp/eisy-aisudo chmod -R 777 /tmp/eisy-aiWith kind regards,Michel
October 26Oct 26 Thanks, yes I realized I had to put in some money, gave it $20 and used almost $1 tonight creating leak sensor monitors for new ones I installed today.Worked very well, but did make mistakes.
October 26Oct 26 Author @Jimbo.Automates ,Excellent. Next release has better usage of tokens.Next step (if you wish):Ask it to create a vscode dev environment for you. Then, connect to it using cursor or vscode (remote), and step through the code. You can start/stop the plugin from vscode/cursor and run step through the code right from your Windows computer.With kind regards,MIchel
October 26Oct 26 Trying to use the AI as a user (not developer) - it seems like it does not like to go back and check PG3 nodes before they gets migrated It is also confused about messaging - but that is kind of ok as it is not being supported going forward (I still like getting a daily summary email) I ran into some issues migrating - so I need to focus on that now
October 27Oct 27 Author @Panda88 ,eisy-ai has absolutely no notion of PG3. Just like the Admin Console, PG3 is going away, and there's absolutely no point in supporting it via eisy-ai and waste precious tokens.With kind regards,Michel
October 27Oct 27 understood, but the transition to the new solution is not smooth - I have reversed a few installed nodes back as they do not work correctly under the new system
October 27Oct 27 Author If true, you must submit a ticket WITH ALL THE LOGS. Just asking "What Can I do ?" requires for us to have developed magical powers of reading your mind and reading your logs remotely. Alas, we have not yet developed that acumen. With AI, maybe soon With kind regards,Michel
October 27Oct 27 Having some nice chats with eisy-ai now! I tried to have it look at a history of my devices, things like energy use and temperature, it said it has no history of such things. How do I enable that? Also surprised that just refreshing the browser it lives in made it forget the conversation we were having. Also it likes to highlight some data in white boxes with white text on top . Impossible to see the text.No complaints, I see this will be very powerful!
October 27Oct 27 Author 3 hours ago, photogeek54 said:Having some nice chats with eisy-ai now! I tried to have it look at a history of my devices, things like energy use and temperature, it said it has no history of such things. How do I enable that? Also surprised that just refreshing the browser it lives in made it forget the conversation we were having. Also it likes to highlight some data in white boxes with white text on top . Impossible to see the text.No complaints, I see this will be very powerful!Please see: New and Improved eisy-ai
October 27Oct 27 Just embarked on an eisy ai plugin creation. It worked quite well. There is now a plugin for Araknsis AN-310-SW-24-POE that monitors the status of all 24 ports and lets you control poe. I suppose I'll upload it to the store. I would feel guilty charging for it seeing as how I did almost nothing. It was scary how easy this was. I have very little to complain about. It required minimal input from me aside from giving it a username and password to work with and directing it toward what functionality I was hoping for. The only annoying thing is the output keeps scrolling the screen. I try to read what it has output to me and the screen keeps jumping down as eisy ai adds more stuff so I just have to quit reading until it decides it is done adding output, which is hard to tell. I never so much as saw a single line of code let alone read or modify anything.I am going to ask it if it thinks any further additions to the plugin would make sense.First impression is very good. This is going to blow up the number of plugins. Obviously user interaction will need to be much more on something more nuanced as to how it should behave.
October 27Oct 27 Author @apostolakisl ,Thank you very much for the feedback, and I am so very glad to hear!Regarding the store, the whole point is to enable developers to develop plugins easily, and let them reap the benefits by supporting it. So, please do not feel guilty as this is just a tool specifically for that purpose. Again, if you are willing to support it, you're welcome to charge for it, and it's totally up to the customers to decide whether or not they wish to pay for it.Regarding code, the new version [ link ] has a file viewer so that you can view the directory, see changes in real time, and edit them. With kind regards,Michel
October 27Oct 27 Author @apostolakisl , by the way, and just for reference, this is how much you spent. Sonnet is the model used for plugin development, and Haiku is for everything else.
October 27Oct 27 20 minutes ago, Michel Kohanim said:@apostolakisl ,Thank you very much for the feedback, and I am so very glad to hear!Regarding the store, the whole point is to enable developers to develop plugins easily, and let them reap the benefits by supporting it. So, please do not feel guilty as this is just a tool specifically for that purpose. Again, if you are willing to support it, you're welcome to charge for it, and it's totally up to the customers to decide whether or not they wish to pay for it.Regarding code, the new version [ link ] has a file viewer so that you can view the directory, see changes in real time, and edit them.With kind regards,MichelOK. Anyway, 99% was really easy, but working on some tweaks is proving challenging. First off, it seems eisy ai forgot it authored the plugin and didnt' know the file path. I had to tell it the file path. I asked it to change the default icons, and it seems convinced it can't be done. And finally the scrolling thing is going to give me a seizure! It bounces around all crazy like while it is outputting. Are others having this experience?EDIT: One last thing, I don't believe there is anyway to give eisy ai screen shots. Am I missing something? Edited October 27Oct 27 by apostolakisl
October 27Oct 27 Author lists my generated plugins - or go to the plugin page in the UI and then say: this pluginLoad plugin x - same as above. The UI page provides context to everything in eisy-ui. So, when you say "this" it'll check the UI contexticons for nodes are unfortunately static and cannot be changed except for those listed. This is going to change soonYes, I have the same issue with scrolling ... it should be fixedBefore you go any further, please make sure you install the new version and copy the files to the new location
October 27Oct 27 3 minutes ago, Michel Kohanim said:lists my generated plugins - or go to the plugin page in the UI and then say: this pluginLoad plugin x - same as above. The UI page provides context to everything in eisy-ui. So, when you say "this" it'll check the UI contexticons for nodes are unfortunately static and cannot be changed except for those listed. This is going to change soonYes, I have the same issue with scrolling ... it should be fixedBefore you go any further, please make sure you install the new version and copy the files to the new locationI couldn't seem to find the method to install eisy ai, and perhaps what I did is less than ideal. I found nucore in git hub and installed it from there and then had to mess around figuring out how to get it to load and stay alive. I installed it from github just on Saturday, has it changed since then?
Create an account or sign in to comment