- Identify why OOM occurs in agents and long-running workflows.
- Separate context, persistent state, logs, and heavy results.
- Apply limits, queues, and alerts before the system dies without explanation.
Common causes
- Storing complete tool responses in the agent state.
- Putting entire documents into context instead of chunks and references.
- Retries that duplicate messages, results, and traces.
- n8n workflows running in a single process without separate workers.
- Huge buffers instead of streaming or batch processing.
politica_memoria:
max_tool_result_chars: 6000
guardar_en_contexto:
- resumen
- ids
- citas
guardar_fuera:
- html completo
- pdf extraido
- respuestas api largas
si_supera_limite:
- persistir_blob
- resumir
- enlazar_por_id
- continuar