Local Ingestion, Cloud Retrieval
OpenCone is not a blanket offline app. It processes files on device, then uses OpenAI and Pinecone to power the core search path.
OpenCone is a native SwiftUI client that extracts and normalizes documents locally, stores credentials in the Keychain, and uses OpenAI plus Pinecone as the cloud execution path for embeddings, vector search, and answer synthesis. It is useful when you want a native Apple front end over your own OpenAI and Pinecone accounts.
What Makes OpenCone Different
Local document preparation
Security-scoped bookmarks, PDFKit parsing, and Vision OCR are used on device to prepare PDFs, images, and other file types before any cloud call is made.
MIME-aware chunking
Text is chunked according to document type and processing stage, rather than passed through a one-size-fits-all ingestion path.
OpenAI embeddings and responses
OpenAI handles the embedding and answer-generation path, with support for modern embedding and response models instead of a fully local inference stack.
Pinecone serverless namespaces
Pinecone indexes and namespaces hold the remote vector search layer, making it easy to separate corpora without running your own vector database service.
Grounded search and answer flow
Queries retrieve relevant chunks from Pinecone, then package that evidence back into an OpenAI response path for grounded answers.
Operational visibility
Processing logs, timings, and status views help users understand which work happened locally and which work required the cloud services behind the app.
Bring-your-own-keys setup
OpenCone assumes the user brings their own OpenAI and Pinecone credentials rather than hiding those dependencies behind a managed backend.
Technical Profile
Core APIs
- Local: SwiftUI, PDFKit, Vision OCR, security-scoped bookmarks
- Embeddings: OpenAI Embeddings API
- Generation: OpenAI Responses API
- Vector search: Pinecone Serverless indexes and namespaces
Data boundary
- Local: files, OCR output, settings, API keys in Keychain
- Remote: embedding requests, vector upserts, search, answer generation
- Truth: cloud-hybrid, not a fully offline RAG stack
- Audience: users who want a native front end over OpenAI and Pinecone
How OpenCone Works
Launch & Setup
The app starts with a bring-your-own-keys flow for OpenAI and Pinecone, storing those credentials locally for subsequent requests.
Add & Process Documents
Files are imported locally, parsed with PDFKit or OCR when needed, chunked on device, and prepared for remote indexing.
Semantic Search
Search requests hit Pinecone for vector retrieval and return the highest-value evidence chunks for answer construction.
Monitor & Configure
OpenAI responses are used to synthesize the final answer, while local logs and metrics keep the full pipeline visible.