Alchimista

Alchimista is an advanced self-hosted solution designed for AI governance, document intelligence, and privacy assurance. This infrastructure layer facilitates the ingestion of evidence, the construction of a tenant-scoped retrieval index, and the generation of mandatory citations. Moreover, it records AI decisions along with reliable privacy metadata to enhance transparency and compliance.

The platform's architecture is designed to operate without reliance on external services such as Google Cloud Platform (GCP) or Auth0, eliminating the need for complex project setups or paid services. However, optional integrations with GCP components like Vertex and Pub/Sub are available for enhanced functionality.

How Alchimista Works

Alchimista employs sophisticated mechanisms to transform local documents into privacy-protected, auditable, and cited responses. It allows users to manage documents securely within its local application framework, which includes functionalities for handling governance and auditing.

Features

• Local Hosting: Alchimista can be deployed entirely on local hardware without external dependencies.
• Privacy Protection: The platform incorporates privacy detection methods; however, it does not certify compliance with GDPR or the AI Act, and its PII (Personally Identifiable Information) detectors may not capture every sensitive value.
• Document Management: Users can ingest documents, maintain a persistent library, and manage auditing evidence all within one local interface.

Usage Examples

To facilitate document ingestion, Alchimista provides a minimal API for integration:

curl -fsS -X POST http://127.0.0.1:8011/v1/ingest \
  -H "Authorization: Bearer ${ALCHIMISTA_API_TOKEN}" \
  -F tenant=default \
  -F doc_id=example-001 \
  -F file=@example.txt

Then, to query the ingested documents with citations:

curl -fsS -X POST http://127.0.0.1:8013/v1/query \
  -H "Authorization: Bearer ${ALCHIMISTA_API_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{"tenant":"default","query":"What does the evidence say?","top_k":3,"doc_ids":["example-001"]}'

Local Runtime Topology

Alchimista's architecture features a streamlined local runtime topology:

browser ──> dashboard ───┬──> ingestion ──direct HTTP──> processor
                         ├──> RAG                         │
                         └──> privacy-service <───────────┘
                                   │
                                   └──> Rizzo model runtime (idle by default)
                         services ─────> PostgreSQL

This structure provides clear navigating pathways from user interactions through to data processing and storage, ensuring efficient handling of privacy policies and audit trails.

Privacy Policies

Alchimista allows nuanced configuration of privacy settings, supporting multiple policies to detect, protect, or simply log sensitive data based on the operational needs:

Conclusion

Alchimista aims to empower organizations with comprehensive tools for AI governance and document intelligence while prioritizing privacy and compliance. Its self-hosting capabilities, coupled with a flexible architecture, make it a valuable asset for managing sensitive information effectively and transparently.