Storage
Org-scoped storage
Files, bases, and membership rules—treat uploads as the source of truth for your pipeline.
⚠️ Pre-alpha: This platform is in early development. Features may change.
Build RAG features, not plumbing
Multimodal vector search and Gemini embeddings—upload once, query by meaning, subscribe to live index updates.
Upload images, video, audio, and documents once. Search by embedding similarity across bases—GraphQL queries, subscriptions, and MCP included.
GraphQL subscriptions, vector search, and org-scoped storage—primitives you call from code.
Run topK similarity search over HNSW-indexed Gemini vectors—filter by embed model in GraphQL, scope to a base client-side.
topK · 768d
query FilesWithinDistance(
$vector: [Float!!]
$topK: Int!
$filter: EmbeddingIndexFilter!
) {
querySimilarEmbeddingIndexByEmbedding(
vector: $vector
topK: $topK
by: embedding
filter: $filter
) {
id
model
vector_distance
file { id name }
}
}variables
vector: [768 floats]
topK: 10
filter: { model: { eq: GEM_2_768 } }
response
1Sony WH-1000XM4distance 0.0912
2Bose QuietComfort 45distance 0.1245
3Sennheiser Momentum 4distance 0.1618
768d · topK: 10 · GEM_2_768 · base scoped client-side
GE2-768
768d · Gemini
GE2-1536
1536d · Gemini
Files, bases, and membership rules—treat uploads as the source of truth for your pipeline.
Indexes update in real time as processing completes—no polling for file status or embeddings.
Cosine HNSW search over Gemini Embedding 2 vectors—filter by model and base at query time.
Images, video, audio, text, and PDF—one upload path, one embedding pipeline.
Patterns teams ship on top of raggen.ai—similarity search, retrieval, and IDE context without stitching storage, ETL, and search vendors.
Vector embeddings power semantic search, RAG retrieval, deduplication, and IDE-integrated knowledge access. raggen.ai unifies multimodal embedding, Dgraph storage, and GraphQL vector APIs.
Verified capabilities in the current schema and processing pipeline.
3
768–3072
Live
5 types
GraphQL queries and subscriptions, API keys, and MCP—what you need when a side project goes live.
Point Cursor, Claude Desktop, or any MCP client at your bases—one endpoint, GraphQL-backed tools.
Add raggen as an MCP server and pull context into the editor while you code.
Query org bases and files from Claude over the MCP transport.
Any MCP-compatible client can call the same GraphQL-backed tool surface.
https://mcp.raggen.aiAPI & MCP docsCommon questions about multimodal vector search, GraphQL, and raggen.ai.
Images (PNG, JPEG, GIF, WebP), video (MP4, MOV, WebM), audio (MP3, WAV), plain text, and PDF. Each file is embedded with Gemini Embedding 2 after upload.
Three Gemini Embedding 2 models: GEM_2_768 (768d), GEM_2_1536 (1536d), and GEM_2_3072 (3072d). A file can hold multiple vectors—one per model—without overwriting prior indexes.
Use the querySimilarEmbeddingIndexByEmbedding GraphQL query with a query vector, topK, and EmbeddingIndexFilter (base and model). Results include cosine distance and linked file metadata.
Yes. File, organization, base, and related types use @withSubscription—subscribe to getOrganization or file lists and receive live updates when indexes change.
Yes, for file-level retrieval: upload corpus files to bases, run vector search to find relevant files, and use File.content (extracted text) or storage URLs as LLM context. Chunk-level RAG is not in the schema today.
raggen.ai includes a free tier: 1 million embedding tokens per org per calendar month (file indexing and vector-search queries) and 100 MB of stored media—no credit card required to start. Beyond the free tier, $0.40 per additional million tokens and $0.05 per GB-month storage. We invoice prior-month overage at the start of each month. Add a payment method to unlock 1536/3072-dimension embed models.
https://mcp.raggen.ai — connect Cursor, Claude Desktop, or custom MCP clients to query org bases and files over GraphQL-backed tools.
Pre-alpha: the platform is in early development and features may change. Core flows—upload, embed, search, subscribe—are available for development and evaluation.
No credit card required—monthly token and storage allowance included on every org.
Start on the free tier: 1 million embedding tokens per org each month and 100 MB storage included. Overage at $0.40/1M tokens and $0.05/GB-month only when you exceed the allowance.
Free tier first; overage only if you exceed the allowance
View free tier detailsSpin up a base, upload sources, and query from your app or IDE—same stack from hackathon to prod.
Upload images, video, audio, and documents once. Search by embedding similarity across bases—GraphQL queries, subscriptions, and MCP included.
Upload complete
0 done