DocDuck¶

Multi-provider document indexing + AI retrieval (RAG) for your internal knowledge. Index OneDrive, S3, or local files; ask natural language questions; get cited answers.

Key Features¶

Multi-provider: OneDrive (business & personal), S3, local filesystem (extensible)
Smart idempotent indexing (ETag tracking, orphan cleanup, force reindex)
Pluggable text extraction (DOCX, TXT/MD, PDF*, ODT, RTF)
Configurable chunking & OpenAI embeddings
Minimal Query API with /query, /chat (streaming), /docsearch & provider filtering
PostgreSQL + pgvector for similarity search
Kubernetes & Docker friendly (CronJob indexer, long-running API)
Secure admin operations (seeded admin, secret-based token)
Pragmatic modern .NET 8 codebase (records, DI, logging)

*PDF requires optional dependency.

Quick Start¶

If you just want to try it locally:

Provision PostgreSQL with pgvector extension
Set environment variables (OpenAI API key, DB connection string)
Run the Indexer once
Query the API

See Quick Start for copy-paste steps.

Documentation Map¶

Audience	Start Here
Casual evaluator	Quick Start
Power user / operator	Installation, Configuration
Architect / engineer	Architecture
Extending providers	Provider Framework
RAG internals	Search & RAG
AI Agent embedding	AI Agent Context

High-Level Architecture¶

Providers → Indexer Pipeline → PostgreSQL (chunks + metadata) ← Query API (RAG) ← User

Indexer runs on a schedule (or ad-hoc) and updates embeddings
Query API performs semantic search + synthesis across stored chunks

See Architecture for detailed diagrams.

Why DocDuck?¶

Production-focused from the start (idempotency, cleanup, metrics-friendly logging)
Lean: only the moving parts needed for reliable RAG over your documents
Extensible: add new providers & text extractors with small focused interfaces
Transparent: clear data model & SQL; easy to audit

Status¶

Active development. API surface & schema considered stable for initial OSS release (v1). Expect additive enhancements.

License¶

MIT — see License.