Add Hubris integration page

integrations/hubris.md

@@ -0,0 +1,118 @@
+---
+layout: integration
+name: Hubris
+description: Use 500+ models from OpenAI, Anthropic, Google, DeepSeek, Qwen and others through Hubris, an OpenAI-compatible LLM gateway billed in Russian rubles
+authors:
+    - name: Hubris
+      socials:
+        github: Aimagine-life
+pypi: https://pypi.org/project/haystack-ai/
+repo: https://github.com/deepset-ai/haystack
+type: Model Provider
+report_issue: https://github.com/deepset-ai/haystack/issues
+logo: /logos/hubris.png
+version: Haystack 2.0
+toc: true
+---
+
+### **Table of Contents**
+
+- [Overview](#overview)
+- [Usage](#usage)
+
+## Overview
+
+[Hubris](https://hubris.pw) is an OpenAI-compatible LLM gateway: one API key and one balance (billed in Russian rubles) for models from OpenAI, Anthropic, Google, DeepSeek, Qwen, Z.ai, Moonshot, xAI, MiniMax and others. The full catalog with prices is at [hubris.pw/models](https://hubris.pw/models); model ids always use the `vendor/model` form (for example `anthropic/claude-sonnet-5`).
+
+## Usage
+
+The Hubris API is OpenAI compatible, so it works with Haystack's OpenAI components out of the box. Create an API key at [hubris.pw/keys](https://hubris.pw/keys) and set it as the `HUBRIS_API_KEY` environment variable.
+
+### Using `ChatGenerator`
+
+```python
+from haystack.components.generators.chat import OpenAIChatGenerator
+from haystack.dataclasses import ChatMessage
+from haystack.utils import Secret
+
+generator = OpenAIChatGenerator(
+    api_key=Secret.from_env_var("HUBRIS_API_KEY"),
+    api_base_url="https://api.hubris.pw/v1",
+    model="anthropic/claude-sonnet-5",
+)
+
+messages = [
+    ChatMessage.from_system("You are a helpful assistant."),
+    ChatMessage.from_user("What is the capital of France?"),
+]
+response = generator.run(messages=messages)
+print(response["replies"][0].text)
+```
+
+### In a pipeline
+
+Here's a question-answering pipeline over a web page. Any model from the [catalog](https://hubris.pw/models) can be used; `google/gemini-3.7-flash` is a good fit for long pages.
+
+```python
+from haystack import Pipeline
+from haystack.components.builders import ChatPromptBuilder
+from haystack.components.converters import HTMLToDocument
+from haystack.components.fetchers import LinkContentFetcher
+from haystack.components.generators.chat import OpenAIChatGenerator
+from haystack.dataclasses import ChatMessage
+from haystack.utils import Secret
+
+template = [
+    ChatMessage.from_user(
+        """According to the contents of this website:
+{% for document in documents %}
+  {{document.content}}
+{% endfor %}
+Answer the given question: {{query}}
+Answer:"""
+    )
+]
+
+pipeline = Pipeline()
+pipeline.add_component("fetcher", LinkContentFetcher())
+pipeline.add_component("converter", HTMLToDocument())
+pipeline.add_component("prompt", ChatPromptBuilder(template=template, required_variables=["documents", "query"]))
+pipeline.add_component(
+    "llm",
+    OpenAIChatGenerator(
+        api_key=Secret.from_env_var("HUBRIS_API_KEY"),
+        api_base_url="https://api.hubris.pw/v1",
+        model="google/gemini-3.7-flash",
+    ),
+)
+
+pipeline.connect("fetcher.streams", "converter.sources")
+pipeline.connect("converter.documents", "prompt.documents")
+pipeline.connect("prompt.prompt", "llm.messages")
+
+result = pipeline.run(
+    {
+        "fetcher": {"urls": ["https://hubris.pw/docs"]},
+        "prompt": {"query": "How do I authenticate requests to Hubris?"},
+    }
+)
+print(result["llm"]["replies"][0].text)
+```
+
+### Embeddings
+
+Embedding models from the catalog work through the OpenAI embedders the same way:
+
+```python
+from haystack.components.embedders import OpenAITextEmbedder
+from haystack.utils import Secret
+
+embedder = OpenAITextEmbedder(
+    api_key=Secret.from_env_var("HUBRIS_API_KEY"),
+    api_base_url="https://api.hubris.pw/v1",
+    model="openai/text-embedding-3-small",
+)
+print(embedder.run(text="Haystack pipelines with Hubris")["embedding"][:5])
+```
+
+Streaming (`streaming_callback`), tool calling (`tools`) and structured output (`response_format`) work exactly as with the stock OpenAI components; the `supported_parameters` field of `GET https://api.hubris.pw/v1/models` lists what each model accepts.