feat: add Valkey vector store integration

[daric93]

Implement ValkeyStore in application/vectorstore/ using valkey-glide-sync and the valkey-search module for HNSW vector similarity search.

Fixes: #2475

What kind of change does this PR introduce?

Feature — adds Valkey as a new vector store backend option.

Why was this change needed?

DocsGPT users who run Valkey as their primary data store had no way to use its vector search capabilities within DocsGPT, requiring a separate vector database. This integration provides a high-performance, low-latency vector search option using Valkey's native HNSW indexing.

Changes

• Add ValkeyStore class extending BaseVectorStore with full interface: search, add_texts, delete_index, get_chunks, add_chunk, delete_chunk
• Register "valkey" in VectorCreator factory
• Add VALKEY_* settings to Settings class (host, port, password, tls, index_name, prefix, request_timeout, distance_metric, vector_type, vector_algorithm)
• Add valkey-glide-sync==2.3.1 to requirements.txt
• Add Valkey config examples to .env-template
• Add 56 unit tests (mocked, no external deps)
• Add 7 integration tests (requires running Valkey with search module)

Other information

• Uses the synchronous GLIDE client (valkey-glide-sync) — consistent with all other vector store implementations in the project. The Flask backend and Celery workers are synchronous — no async runtime exists in the application.
• Requires Valkey server with the valkey-search module loaded
• Set VECTOR_STORE=valkey to enable
• Tested against Valkey 8.1 / 9.1 with valkeysearch module
• Soft import guard prevents breaking other vector store backends when valkey-glide-sync is not installed
• Pagination uses a shared generator with a safety cap of 1000 pages (~10M documents) to prevent infinite loops
• Search k is capped at [1, 100] to prevent server memory exhaustion
• delete_index re-raises exceptions so callers can handle failures during re-indexing
• Supports context manager (with ValkeyStore(...) as store:) for deterministic connection cleanup

[vercel]

@daric93 is attempting to deploy a commit to the Arc53 Team on Vercel.

A member of the Team first needs to authorize it.

[daric93]

Hi @dartpain

I'd appreciate a review on this when you get a chance — if Valkey vector store backend aligns with the project's roadmap. If someone else is better suited to review the storage/vector store layer, could you point me their way?

Thanks!

[atao2004]

Hey @daric93 👋

I've opened a PR against your feat/valkey-vector-store branch that adds client_name="docsgpt_vector_store_client" to both GlideClientConfiguration paths in valkey.py: daric93#2

This sends a CLIENT SETNAME command on connection, making DocsGPT connections identifiable in CLIENT LIST and monitoring dashboards. One-line addition per config path, zero behavioral changes.