Production Best Practices

This guide covers recommended patterns for running Papr in production workloads.

1) Scope Every Request Explicitly

For multi-tenant products, include:

• organization_id
• namespace_id
• external_user_id

Apply this to both writes and reads.

2) Standardize Policy Defaults

Set default memory_policy conventions:

• mode: auto for unstructured, manual for structured imports
• consent: based on your legal model
• risk: explicit risk tiering
• acl: deterministic read/write templates

3) Tune Retrieval for Quality

Baseline recommendations:

• enable_agentic_graph=true
• max_memories=15-20
• max_nodes=10-15

Increase limits for deeper multi-hop use cases.

4) Use Messages Compression for Long Sessions

For chat-heavy systems:

• Store all turns via /v1/messages
• Use /compress and context_for_llm for prompt budgets

5) Add Feedback Loop

Capture retrieval quality signals through /v1/feedback to improve quality tracking and regression detection.

6) Operational Reliability

• Add retries with bounded backoff.
• Log request IDs and memory/search IDs.
• Protect write paths with idempotency checks in your ingestion layer.

7) Portability and Backup

Use OMO export/import for:

• migration,
• archival,
• interoperability with OMO-compatible systems.

Related

• Troubleshooting
• Migration Guide