add origin_power_ai plugin v0.0.5 - #3022
Conversation
…ields, min dify version, .git excluded from package)
|
Friendly ping 🙏 This PR adds the origin_power_ai model provider plugin (v0.0.5, reported as ready for review: manifest Would appreciate a review from the team when there is bandwidth — happy to address any feedback. Thank you! |
|
Closing this PR because the current Pre Check Plugin / pre-check-plugin CI run failed. CI summary:
Please address the reported failures (or rerun CI if the failure was transient) before requesting reopening or resubmitting. |
|
All three reported failures have been addressed in the meantime (the CI run predates these fixes):
A follow-up commit (17a5315 updating the README) is also on the PR branch, so a re-run of Could you please reopen the PR? Happy to iterate on any further feedback. Thank you! 🙏 |
Plugin Submission
Plugin information
Submission type
What changed
Adds the Origin Power AI model provider plugin. Origin Power AI is a unified OpenAI-compatible gateway that aggregates GLM, MiniMax and other mainstream models behind a single API key.
openai_api_compatibleplugin, Apache-2.0)PRIVACY.mdbundled; manifestrepo,contactandmeta.minimum_dify_versionfields setRisk level
The plugin forwards user prompts and conversation context to the API endpoint configured by the workspace administrator (an arbitrary user-supplied URL) via plain HTTP requests. No other external calls; nothing is sent to the plugin author.
Local validation
.gitexcluded, all bundled YAMLs parse,_position.yamlmatches the predefined model set).GET /v1/models, and a real end-to-end chat invocation (streaming) passed.Plugin test result: success).Security and privacy notes
The plugin makes network requests only to the model gateway endpoint configured by the workspace administrator when setting up the provider. This is the arbitrary-URL capability detected by the scanner:
requests.getinprovider/origin_power_ai.py(credential validation viaGET /models) andrequests.postinmodels/llm/llm.py(OpenAI-compatible/chat/completionscalls). Requests carry only the conversation data required to call the LLM API and are authenticated with the API key entered by the administrator. The plugin does not call any other service, does not send data to the plugin author, and does not persist data locally.Required checks