Add support for Qdrant Cloud Inference as an embedding provider

[bashandbone]

Qdrant client and Qdrant Cloud now offer cloud inference with Fastembed, allowing users to utilize the same models while offloading embedding computations to the cloud.

Feature request:

• Integrate Qdrant Cloud Inference as an embedding provider in Codeweaver.
• Users should be able to select Qdrant Cloud Inference as an embedding backend, in addition to currently supported local or self-hosted options.
• Ensure configuration options allow switching easily between local inference and cloud inference.
• Provide documentation for setup and usage, including API credentials, supported models, and usage examples.

Benefits:

• Simplifies scaling: Users can leverage Qdrant's managed infrastructure to handle increased embedding workloads.
• Consistency: Enables users to use the same models irrespective of local or cloud execution.
• Reduced local resource consumption: Cloud inference offloads compute from client machines.

References:

• Qdrant client documentation on Cloud Inference
• Fastembed integration details

Acceptance criteria:

• Qdrant Cloud Inference integration is available and easily selectable.
• End-user documentation is updated and tested.
• Includes basic tests to validate provider selection and inference results.

[github-actions]

Claude finished @bashandbone's task —— View job

---

Issue Analysis: Qdrant Cloud Inference Support

• Gather context and review codebase
• Analyze current embedding provider architecture
• Research Qdrant Cloud Inference API
• Estimate implementation difficulty
• Identify related issues/code
• Provide summary and recommendations

---

Issue Summary

This feature request proposes adding Qdrant Cloud Inference as an embedding provider in CodeWeaver. Qdrant Cloud Inference allows users to offload embedding computations to the cloud while using the same FastEmbed models available locally.

---

Current Architecture Analysis

CodeWeaver already has a well-structured embedding provider system:

Existing Providers (src/codeweaver/providers/embedding/providers/):

• base.py - Abstract EmbeddingProvider base class (line 90)
• fastembed.py - Local FastEmbed provider (lines 82-143)
• Plus: voyage.py, openai_factory.py, cohere.py, bedrock.py, google.py, huggingface.py, mistral.py, sentence_transformers.py

Key Integration Points:

• EmbeddingProvider[EmbeddingClient] generic base class at src/codeweaver/providers/embedding/providers/base.py:90
• Provider configuration via EmbeddingProviderSettings at src/codeweaver/config/providers.py:195
• Existing QdrantVectorStore at src/codeweaver/providers/vector_stores/qdrant.py:26 already uses AsyncQdrantClient
• Provider enum at src/codeweaver/providers/provider.py:64 (would need new QDRANT_INFERENCE entry)

---

Qdrant Cloud Inference API Overview

Based on Qdrant documentation, Cloud Inference provides:

Key Features:

1. Simple client initialization: QdrantClient(..., cloud_inference=True)
2. Document-based embedding: Uses models.Document(text="...", model="model-name")
3. Supported models: Dense models (e.g., sentence-transformers/all-MiniLM-L6-v2) and sparse models (qdrant/bm25)
4. Hybrid search: Supports both dense and sparse vectors simultaneously

API Pattern:

from qdrant_client import QdrantClient, models

client = QdrantClient(
    url="https://your-cluster.qdrant.io",
    api_key="<your-api-key>",
    cloud_inference=True,
)

# Embedding via Document object
doc = models.Document(text="text to embed", model="sentence-transformers/all-MiniLM-L6-v2")

---

Implementation Approach

Two potential approaches:

Option A: Standalone Embedding Provider (Recommended)

Create a new QdrantCloudInferenceProvider class that:

• Extends EmbeddingProvider[QdrantClient]
• Uses QdrantClient with cloud_inference=True
• Returns embeddings via models.Document objects
• Supports both dense and sparse models

Pros: Clean separation of concerns, follows existing provider patterns
Cons: Requires maintaining separate embedding client from vector store

Option B: Integrated with VectorStore

Enhance QdrantVectorStore to use cloud inference directly during upsert/search operations.

Pros: Simpler integration, fewer moving parts
Cons: Tightly couples embedding to storage, less flexible

---

Related Issues

Issue	Title	Relevance
#110	Fix vector store type usage and implement Qdrant filter conversion	Qdrant improvements
#104	Expand vector provider support and fix capabilities definition	Provider system enhancements
#108	Add LiteLLM, Cerebras, and OpenRouter embedding provider support	Similar embedding provider additions
#93	Convert provider TypedDicts to Pydantic models	Configuration refactoring
#83	Implement dynamic provider registration system	Provider architecture
#116	[EPIC] Replace registries with dependency injection system	Architecture changes may affect implementation

---

Implementation Difficulty Estimate

Difficulty: Medium ⚠️

Effort estimate: 3-5 days for a complete implementation

Complexity factors:

1. ✅ Low: Qdrant client already a dependency
2. ✅ Low: Clear API pattern from qdrant-client documentation
3. ⚠️ Medium: Need to handle async embedding operations (Qdrant Cloud Inference returns embeddings server-side)
4. ⚠️ Medium: Configuration system needs new provider type and settings
5. ⚠️ Medium: May need to coordinate with QdrantVectorStore for optimal usage
6. ⚠️ Medium: Testing requires Qdrant Cloud account or mocking

Implementation tasks:

1. Add Provider.QDRANT_INFERENCE to provider enum
2. Create QdrantCloudInferenceProviderSettings TypedDict
3. Implement QdrantCloudInferenceProvider class
4. Add model capabilities for supported models
5. Update documentation
6. Write unit and integration tests

---

Potential Impact on Existing Functionality

Low risk to existing features:

• New provider won't affect existing FastEmbedEmbeddingProvider or QdrantVectorStore
• Users can continue using local FastEmbed with existing configuration
• Configuration is additive (new settings, not replacing existing)

API considerations:

• The EmbeddingProvider.embed_documents() interface returns list[list[float]]
• Qdrant Cloud Inference typically handles embedding internally during upsert
• May need adapter pattern to extract embeddings for caching/reuse

---

Recommended Labels

• enhancement ✅ (already applied)
• embeddings ✅ (already applied)
• fastembed ✅ (already applied)
• Suggest adding: providers, qdrant

---

Next Steps

1. Validate approach: Confirm Option A (standalone provider) aligns with project architecture goals
2. Check issue [EPIC] DI + Monorepo: Unified architecture transformation #116: DI system changes may affect how providers are registered
3. Prototype: Create basic provider implementation to validate API integration
4. Documentation: Plan docs structure for cloud vs local inference setup

---

References

• Qdrant Cloud Inference Documentation
• Qdrant Client Python README
• Hybrid Search Tutorial
  |