DocuMind API Service • Operational

DocuMind Intelligence API

Document Q&A assistant featuring Retrieval-Augmented Generation (RAG), vector similarity with pgvector, and bounded agent tool execution.

Environment
PRODUCTION
Port & Host
127.0.0.1:8888
LLM Model
gpt-4.1-mini
Embedding Dimensions
1536d

API Endpoints & Documentation

OpenAPI Interactive Docs SWAGGER
Interactive Swagger UI to test and inspect all API endpoints, schemas, and request models directly.
GET /docs →
ReDoc Documentation REDOC
Clean, responsive OpenAPI specification and reference documentation for developer integration.
GET /redoc →
Process Liveness JSON
Lightweight process liveness probe returning instant service status for load balancers and systemd.
GET /health →
System Readiness JSON
Deep readiness probe verifying active PostgreSQL connection pool and pgvector extension availability.
GET /ready →
Application Metadata JSON
Current runtime configuration, supported file limits, active models, and available bounded tools.
GET /api/v1/meta →
Evaluation Benchmark JSON
Empirical benchmark results comparing 300-token and 800-token chunk configurations across gold set questions.
GET /api/v1/evaluation →

Architectural Guardrails

Untrusted Document Scope Document text is quarantined in <untrusted_context> tags. Instructions inside docs are ignored.
Verifiable Citations Citation IDs returned to users are verified against stored candidate chunks to eliminate citation hallucination.
Confirmed Action Writes Ticket creation requires an expiring (10m TTL), single-use confirmation token with idempotency keys.
Hidden Backend Topology Internal reverse proxy hides port 8888. Client requests use same-origin relative endpoints.