CLI Implementation

Internals of the warmup_rag management command. User-facing usage lives in the CLI guide.

File: django_ai_sdk/management/commands/warmup_rag.py. The SDK also ships refresh_integrations (django_ai_sdk/integrations/mcp/management/commands/refresh_integrations.py) for MCP integrations.

Overview

warmup_rag pre-builds RAG indexes for all registered agents and memories so the first chat request doesn’t pay the warmup cost. It is a standard Django BaseCommand that runs async work via asyncio.run().

Arguments

Option	Type	Description
`--agent`	`str`	Only warm up a specific agent (matched by class name)
`--memory`	`str`	Only warm up a specific memory ID
`--force-rebuild`	flag	Rebuild indexes from scratch, deleting existing ones

Implementation Details

Resolving memories

from django_ai_sdk.memories.models import Memory

all_memory_ids = [str(m.id) for m in Memory.objects.all().only("id")]

if memory_filter:
    if memory_filter not in all_memory_ids:
        raise CommandError(f"Memory '{memory_filter}' not found")
    memory_ids = [memory_filter]
else:
    memory_ids = all_memory_ids

With no --memory, every Memory row is warmed. An unknown memory ID raises CommandError.

Ensuring the registry is ready

from django_ai_sdk.agents.registry import registry

try:
    agents = registry.all()
except RuntimeError:
    registry.setup(instantiate=True)
    agents = registry.all()

Filtering by agent

--agent matches the class name, not the UUID:

if agent_filter:
    filtered = {
        aid: inst for aid, inst in agents.items()
        if inst.__class__.__name__ == agent_filter
    }
    if not filtered:
        raise CommandError(f"Agent '{agent_filter}' not found in registry")
    agents = filtered

Warming up

for aid, inst in agents.items():
    if inst.rag_provider is None:
        self.stdout.write(self.style.WARNING(f"[{name}] No RAG provider configured, skipping"))
        continue

    for mid in memory_ids:
        try:
            await inst.rag_provider.warmup(inst, mid, force_rebuild=force_rebuild)
            self.stdout.write(f"  [{name}] Warmed memory {mid[:8]}")
        except Exception as e:
            self.stdout.write(self.style.ERROR(f"  [{name}] memory {mid[:8]} FAILED: {e}"))
            logger.exception("Warmup failed for %s memory %s", name, mid)

Per-memory failures are logged and don’t abort the run. Agents without a rag_provider are skipped. Empty states (no agents, no memories) exit with a warning.

Code Equivalent

from django_ai_sdk.agents.registry import registry

agents = registry.all()
for aid, inst in agents.items():
    if inst.rag_provider is None:
        continue
    for mid in memory_ids:
        await inst.rag_provider.warmup(inst, mid, force_rebuild=False)

Manual Testing

python manage.py warmup_rag --agent PirateBasicAgent --memory <memory-uuid> --force-rebuild
python manage.py warmup_rag --memory not-a-uuid   # CommandError
python manage.py warmup_rag --agent Nope          # CommandError

Test Tooling Settings Reference