Approval issue created: CONTRIB-9696

This plugin gives extensive support to AI integration. Thank you so much and if the later version could provide OpenAI-API-compatible for alternative providers, that will be amazing!

Hi Aaron, thank you for your reply. Basically this already does exist, but we were hesitant to provide the possibility, because there is no such thing as an OpenAI-compatible API. They all at least to some minor extent have their own ways of for example returning different errors etc. So we did not want users to believe that our plugin is buggy, because people are trying to use "OpenAI-compatible APIs" But we're likely to add a switch soon to allow this as we of course see the necessity for this. Thanks for your response!

is this plugin working? I keep getting error 404 in AI chat and tiny AI, other AI doesnt work too. I use Gemini AI first, doesnt work, then use Vertex AI, still doesnt work.

gemini model 1.5 flash has been retired, thats why I keep getting error 404. Just in case someone else has the same issue and is as foolish as I am.

Would be nice if we can type our own model instead of providing dropdown, to prevent further issues in case another model is retired so we can just input our own model.

Hi, I'm trying to probe the plugin with ollama local server and llama3.1 model:
Ollama server receive the request:
l will not be utilized
Oct 07 14:56:06 debian-moodle.nos.localdomain ollama[33671]: llama_context: CPU output buffer size = 0.50 MiB
Oct 07 14:56:06 debian-moodle.nos.localdomain ollama[33671]: llama_kv_cache_unified: CPU KV buffer size = 512.00 MiB
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: llama_kv_cache_unified: size = 512.00 MiB ( 4096 cells, 32 layers, 1/1 seqs), K (f16): 256.00 MiB, V (f16): 256.00 MiB
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: llama_context: CPU compute buffer size = 300.01 MiB
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: llama_context: graph nodes = 1126
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: llama_context: graph splits = 1
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: time=2025-10-07T14:56:07.688+02:00 level=INFO source=server.go:1289 msg="llama runner started in 104.42 seconds"
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: time=2025-10-07T14:56:07.697+02:00 level=INFO source=sched.go:470 msg="loaded runners" count=1
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: time=2025-10-07T14:56:07.699+02:00 level=INFO source=server.go:1251 msg="waiting for llama runner to start responding"
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: time=2025-10-07T14:56:07.701+02:00 level=INFO source=server.go:1289 msg="llama runner started in 104.44 seconds"
Oct 07 14:56:07 debian-moodle.nos.localdomain ollama[33671]: [GIN] 2025/10/07 - 14:56:07 | 200 | 1m45s | 127.0.0.1 | POST "/api/generate"

but it seems that where the ollama server try to send the answer, Moodle fails:
Error code: generalexceptionmessage
* line 190 of /local/ai_manager/classes/local/prompt_response.php: TypeError thrown
* line 69 of /local/ai_manager/tools/ollama/classes/connector.php: call to local_ai_manager\local\prompt_response::create_from_result()
* line 216 of /local/ai_manager/classes/manager.php: call to aitool_ollama\connector->execute_prompt_completion()
* line 82 of /local/ai_manager/classes/external/submit_query.php: call to local_ai_manager\manager->perform_request()
* line ? of unknownfile: call to local_ai_manager\external\submit_query::execute()
* line 253 of /lib/external/classes/external_api.php: call to call_user_func_array()
* line 83 of /lib/ajax/service.php: call to core_external\external_api::call_external_function()

Any hint wil appreciated!

Hi,
it's a bit difficult to debug just from your output. Maybe it's sufficient to raise the timeout in the AI manager admin configuration setting?

Hi, but using the built-in plugin Ollama I receive a correct answer. I've made this plugin work using https://localhost/ollama (served by a reversed procy), but if in local_ai_manager I can't get any response unless using https://localhost/ollama/api/generate. Did you ever try ollama in local_ai_manager?
tnx

I'm not sure I understand. You are supposed to define the whole ollama endpoint which would be https://your_ollama_host/ollama/api/generate. It's maybe different to what you have to insert into the core_ai ollama provider, but the local_ai_manager is not at all related to the core_ai subsystem, so it's implemented differently.

Sorry. probably I haven't explained it well. I used core_ai_subsystem in order to test the ollama server. It works, but I want to use local_ai_manager, and in this case Moodle fails parsing the result sent by the ollama server. So, I need to know the parameters that I must use to setup the connection of local_ai_manager to the ollama server. The model is llama3.1.

It's supposed to work by providing the endpoint in the format https://your_ollama_host/api/generate. That should be sufficient.

ok, thanks. I will set a greater timeout, as you said, this can be the cause.