> ## Documentation Index
> Fetch the complete documentation index at: https://docs.interaone.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Redis messaging and queues

> See how Redis supports BullMQ, AI streams, cache entries, locks, Socket.IO coordination, and background processing.

Redis is InteraOne's coordination layer. It carries work between services, live AI output back to the gateway, short-lived conversation state, and distributed locks.

## Workload map

| Mechanism              | Examples                                                                                                      |
| ---------------------- | ------------------------------------------------------------------------------------------------------------- |
| BullMQ                 | AI processing, document ingestion, conversation analysis, agent assist, email, analytics, subscription expiry |
| Streams/pub-sub        | Ordered AI chunks, tool events, final responses, assist responses                                             |
| Cache                  | Conversation gates and short-lived state used by pipeline prechecks                                           |
| Locks                  | Duplicate URL-ingestion prevention                                                                            |
| Socket.IO coordination | Cross-instance real-time delivery when scaling the gateway                                                    |

## Delivery considerations

Redis-backed processing is asynchronous and retryable. Jobs may wait during an outage and may be delivered again after a worker failure. Use durable Redis persistence, monitor failed jobs, and keep handlers idempotent.

<Warning>
  Treat Redis as sensitive infrastructure. It contains tenant identifiers, message work, and internal events. Require authentication, use private networking, and do not expose port `6379` publicly.
</Warning>

## Capacity signals

* Queue depth and oldest-job age
* Worker throughput and failure rate
* Redis memory, eviction policy, CPU, connections, and command latency
* Stream consumer lag
* Persistence health and restart recovery time
* Lock TTLs and stale lock frequency

For production, size Redis for both queued payloads and transient streams, not only cache keys. An eviction policy that removes active queue data can corrupt processing semantics.