I tried enabling the “embeddings module,” but I got the same error as before: “Sorry, an error has occurred.” Other AI features are running fine, though. Could it be something wrong with my settings, or is there something else I should check?
AI-generated summary
Users are reporting issues with Gemini Embeddings in Discourse AI, encountering errors like “Sorry, an error has occurred” despite other AI features working. joo shared a screenshot of their settings and suspects a configuration problem. sam confirmed the complexity of current embedding setup and shared a detailed error traceback, noting that the team is already working on a revamp to simplify configuration. Falco confirmed the revamp is underway and will be released in the coming weeks. In the meantime, Roman provided a temporary fix via a GitHub pull request (#1019) to resolve the current error. The issue is being tracked as a bug under the AI category.
Yeah, I agree configuring embeddings today is complicated, even when documented, I tried configuring this @Falco / @Roman and got:
NoMethodError (undefined method `instance' for class DiscourseAi::Inference::GeminiEmbeddings)
lib/site_settings/type_supervisor.rb:287:in `validate_value'
lib/site_settings/type_supervisor.rb:168:in `to_db_value'
lib/site_setting_extension.rb:433:in `add_override!'
lib/site_setting_extension.rb:678:in `block in setup_methods'
lib/site_setting_extension.rb:500:in `public_send'
lib/site_setting_extension.rb:500:in `set'
lib/site_setting_extension.rb:513:in `set_and_log'
app/services/site_setting/update.rb:65:in `save'
lib/service/base.rb:148:in `instance_exec'
lib/service/base.rb:148:in `call'
lib/service/base.rb:443:in `block in run!'
lib/service/base.rb:443:in `each'
lib/service/base.rb:443:in `run!'
lib/service/base.rb:436:in `run'
lib/service/base.rb:295:in `call'
lib/service/runner.rb:146:in `result'
lib/service/runner.rb:131:in `call'
lib/service/runner.rb:126:in `call'
lib/service/base.rb:296:in `call'
The good news though is that we are working on a revamp of all our embedding config, moving this to a bug report.
Revamp work is undergoing and landing in the next weeks, will update here when it’s ready.
Things will definitely get better once we revamp embedding config, but in the meantime, this should fix the issue:
