Add eval harness (#62 P4), status frames (#96), and agentic runs (#63).

Ship the Phase 4 accuracy eval suite with a manual Gitea workflow, emit
versioned WS status frames during grounded chat, and introduce opt-in
agent infrastructure (Redis/Celery, AgentRun/Step, tools, orchestrator)
gated by ALLOW_AGENTIC_TASKS so default chat behaviour stays unchanged.

.env.prod.example

@@ -101,6 +101,21 @@ FRONTEND_BASE_URL=https://chat.aimloperations.com
 # STRIPE_CHECKOUT_CANCEL_URL=https://chat.aimloperations.com/billing/cancel
 # STRIPE_PORTAL_RETURN_URL=https://chat.aimloperations.com/account/
 
+# Agentic task execution (#63). Keep false until Redis/Celery worker + Ollama
+# capacity are confirmed on this host; false = identical behavior to #62.
+ALLOW_AGENTIC_TASKS=false
+# Shared Redis (channel layer fan-out across gunicorn/uvicorn workers +
+# Celery broker for agent runs). Point both at the same instance.
+# REDIS_URL=redis://10.0.0.128:6379/0
+# CELERY_BROKER_URL=redis://10.0.0.128:6379/0
+# OLLAMA_MODEL_ORCHESTRATOR=gpt-oss:20b
+# OLLAMA_MODEL_SUBAGENT=llama3.2
+# AGENT_MAX_PLAN_STEPS=8
+# AGENT_MAX_ITERATIONS=12
+# AGENT_WALL_CLOCK_SECONDS=600
+# AGENT_SUBAGENT_CONCURRENCY=3
+# Start the worker (server-infra): docker compose --profile agentic up -d worker
+
 # Gunicorn / ASGI (UvicornWorker for WebSockets)
 GUNICORN_WORKERS=2
 GUNICORN_BIND=0.0.0.0:8000