File System Integration

What you'll accomplish: Connect agents to document processing pipelines and S3/object storage.

Document processing

For claims documents, policy PDFs, and other files:

# Install OCR / document processing skill
hoziron catalog install document-ocr

# Mount a document directory (Docker)
docker run -v /mnt/claims-docs:/data/documents:ro ...

The agent can reference documents by path within its allowed directory scope.

S3 / Object storage

hoziron catalog install s3-connector
hoziron vault set AWS_ACCESS_KEY_ID
hoziron vault set AWS_SECRET_ACCESS_KEY
hoziron vault set S3_BUCKET_NAME
hoziron integration start s3-connector

Permission model

Skills are gated by the permission system. A competency declares what it needs:

# In COMPETENCY.toml
required_permissions = ["filesystem:read", "s3:read"]

Agents can only access file paths and buckets permitted by their equipped competency.

PII boundaries

Data flowing between agents in a workflow passes through PII tokenization:

  • Outbound data is tokenized: real PII replaced with tokens
  • Each agent's memory scope is isolated
  • Sensitive customer data (policy numbers, SSNs) never leaks across agent boundaries

Next steps

Related: