v0.1.0 PoC Build: Passing Rust Edition

Stop Guessing.
Start Observing.

The API-first research platform where every chunk, retrieval, and reasoning step is fully observable. Evidence-based answers for production-grade AI.

"Trustworthy answers you can verify, debug, and defend."

Most RAG systems are black boxes. You feed in documents, get answers, and hope they're right. When compliance asks how the AI reached a conclusion, you shrug.

Apex Accelerator changes that. Every answer ships with an Evidence Pack. Every decision is traced in Grafana and Langfuse.

// Evidence Pack v1.2

"claim": "Product supports OIDC",

"source_id": "doc_882",

"uri": "s3://vault/security.pdf",

"confidence": 0.98,

"validator": "PASSED"

Engineered for Production

Robust features for developers who need more than a Jupyter notebook.

Hybrid Retrieval

Dense (vector) + Sparse (BM25) + FTS with RRF fusion. 4 Search modes including Auto-selection.

Full Observability

OpenTelemetry traces, Prometheus metrics, and Langfuse integration for end-to-end lineage.

Multi-tenant Isolation

Tenant-scoped caching and cross-tenant access prevention via Keycloak OIDC.

Multi-agent Graphs

Structured query classification with 6 specialized tools including SQL allowlisting.

Advanced Chunking

9 strategies including semantic (LLM-guided) and recursive, with Tesseract OCR fallback.

Evaluation Framework

Recall@k, MRR, and NDCG metrics. Red-team tested with 95%+ pass rate on adversarial injections.

Trace Explorer

Click a step to see OpenTelemetry span details

LIVE FEED
Select a trace component to inspect OTel span data

The Apex Stack

Rust PostgreSQL Qdrant OpenTelemetry Keycloak Docker OpenAI