feat: Add EmpirioLabs model provider plugin

[Adam-Dalloul]

Summary

Adds the EmpirioLabs model provider plugin under models/empiriolabs/.

EmpirioLabs (EmpirioLabs AI) is an OpenAI-compatible API that serves frontier chat and embedding models through one endpoint, https://api.empiriolabs.ai/v1, with a public GET /v1/models catalog. This plugin mirrors the CometAPI provider plugin and extends the OpenAI-compatible base model classes (OAICompatLargeLanguageModel and the OpenAI-compatible text embedding model), so configuration, streaming, tool calling, and reasoning work the way Dify users already expect from an OpenAI-compatible provider.

It ships 7 predefined chat models plus 1 embedding model, and also supports the customizable-model option so any other slug from the EmpirioLabs catalog can be added by name:

• LLM: qwen3-7-plus, qwen3-7-max, deepseek-v4-pro, deepseek-v4-flash, glm-5-1, kimi-k2-7-code, minimax-m3
• Text embedding: text-embedding-v4

Provider links: website https://empiriolabs.ai, docs https://docs.empiriolabs.ai, API keys https://platform.empiriolabs.ai/dashboard/api-keys

Change Type

• Documentation / non-plugin change
• Non-LLM plugin (tools, extensions, datasource, etc.)
• LLM plugin

LLM Plugin Checklist

Areas affected by this change (check all that apply)

• Message flow (system messages, user to assistant turn-taking)
• Tool interaction flow (multi-round usage, Agent App and Agent Node)
• Multimodal input (images, PDFs, audio, video, etc.)
• Structured output (JSON, XML, etc.)
• Token consumption metrics
• Other LLM functionality (reasoning, grounding, prompt caching, etc.)
• New models / model parameter fixes

Version

• Top-level version in manifest.yaml is 0.0.1 (new plugin)
• dify_plugin is declared in pyproject.toml and locked in uv.lock

Testing

• Validated locally: every model YAML parses against dify_plugin.AIModelEntity, the provider config parses against ProviderEntity, all Python modules compile, and the predefined globs, model sources, position file, and icon assets all resolve.

[gemini-code-assist]

Code Review

This pull request introduces the EmpirioLabs model provider plugin for Dify, adding support for various frontier LLMs and text embedding models. The review feedback highlights several critical improvement opportunities: copying model_parameters before mutation to prevent side effects, aligning the thinking_budget parameter name with the configuration, decoding token slices back to strings before invoking the embedding API to ensure compatibility, renaming a copy-pasted class name, and adding defensive checks to prevent potential AttributeError and KeyError exceptions when handling responses and credentials.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Passing raw token lists (list[list[int]]) directly to the embedding API violates the Union[list[str], str] type hint and may cause 400 Bad Request errors on OpenAI-compatible endpoints that do not support token ID inputs. Decoding the token slices back to strings before invoking the API ensures maximum compatibility and type safety.

Suggested change

	embeddings_batch, embedding_used_tokens = self._embedding_invoke(
	model=model,
	client=client,
	texts=tokens[i : i + max_chunks],
	extra_model_kwargs=extra_model_kwargs,
	)
	chunk_texts = [enc.decode(t) for t in tokens[i : i + max_chunks]]
	embeddings_batch, embedding_used_tokens = self._embedding_invoke(
	model=model,
	client=client,
	texts=chunk_texts,
	extra_model_kwargs=extra_model_kwargs,
	)

[gemini-code-assist]

The model_parameters dictionary is mutated inside _invoke by popping elements. Since dictionaries are passed by reference in Python, this can lead to unexpected side effects if the caller reuses the same parameters (e.g., during retries or multi-agent execution). Copying the dictionary first prevents these side effects.

Suggested change

	self._update_credential(model, credentials)
	self._update_credential(model, credentials)
	model_parameters = model_parameters.copy()

[gemini-code-assist]

The parameter name defined in glm-5-1.yaml is thinking_budget, but llm.py pops reasoning_budget. This mismatch means thinking_budget will not be popped and will be sent directly to the API as an unknown top-level parameter, potentially causing a 400 Bad Request error. Popping both ensures compatibility.

Suggested change

	reasoning_budget = model_parameters.pop('reasoning_budget', None)
	reasoning_budget = model_parameters.pop('thinking_budget', None)
	if reasoning_budget is None:
	reasoning_budget = model_parameters.pop('reasoning_budget', None)

[gemini-code-assist]

The class name OpenAITextEmbeddingModel appears to be a copy-paste leftover from the OpenAI plugin. It should be renamed to EmpirioLabsTextEmbeddingModel to maintain consistency and avoid potential confusion or naming conflicts.

Suggested change

	class OpenAITextEmbeddingModel(_CommonOpenAI, TextEmbeddingModel):
	class EmpirioLabsTextEmbeddingModel(_CommonOpenAI, TextEmbeddingModel):

[gemini-code-assist]

If resp.usage is None (e.g., due to an empty response or API error), accessing resp.usage.prompt_tokens will raise an AttributeError. Adding a defensive check prevents potential crashes.

Suggested change

	usage=self._calc_response_usage(
	model=model,
	credentials=credentials,
	prompt_tokens=resp.usage.prompt_tokens,
	completion_tokens=resp.usage.completion_tokens,
	),
	usage=self._calc_response_usage(
	model=model,
	credentials=credentials,
	prompt_tokens=resp.usage.prompt_tokens if resp.usage else 0,
	completion_tokens=resp.usage.completion_tokens if resp.usage else 0,
	),

[gemini-code-assist]

If credentials is empty or missing the api_key, a KeyError will be raised during validation, resulting in an unhandled traceback and a 500 error. Adding a defensive check and raising CredentialsValidateFailedError provides a clean, user-friendly error message.

Suggested change

	try:
	model_instance = self.get_model_instance(ModelType.LLM)
	if not credentials or not credentials.get("api_key"):
	raise CredentialsValidateFailedError("API key is required")
	try:
	model_instance = self.get_model_instance(ModelType.LLM)

[gemini-code-assist]

Accessing credentials['api_key'] directly can raise a KeyError if the key is missing. Using .get() is safer and adheres to defensive programming best practices.

Suggested change

	credentials_kwargs = {
	"api_key": credentials['api_key'],
	"timeout": Timeout(315.0, read=300.0, write=10.0, connect=5.0),
	"max_retries": 1,
	}
	credentials_kwargs = {
	"api_key": credentials.get('api_key', ''),
	"timeout": Timeout(315.0, read=300.0, write=10.0, connect=5.0),
	"max_retries": 1,
	}