The AI switched off for an audit: switched off is not absent

Goal Switch the hub's AI off without changing edition or losing data, keep reading the sources, and know what to redo when it is switched back on.

A hospital pharmacy sends the hub its dispensing orders every day. The data protection officer asks for the AI to be switched off for the length of an audit.

the operator sets LUMNIK_AI_ENABLED=false and restarts the hub: rag and llm stay installed, said INACTIVE;
ask, search and chat answer the same sentence; the corpus is kept, and the hub says so;
the files are still read; a lifecycle declared during the audit is not indexed, and the apply says so.

lumnik Pro
Hospital pharmacy
Compliance
AI
Operations
Connector
Workflow

dispensing-20261107.csv

order_no;ward;item;status;qty
PH-26-0411;Cardiologie;Amiodarone 200 mg;DELIVREE;60
PH-26-0412;Pédiatrie;Paracétamol sirop 2,4 %;DELIVREE;4
PH-26-0413;Réanimation;Noradrénaline 8 mg/4 ml;PREPAREE;20
PH-26-0414;Oncologie;Ondansétron 8 mg;DELIVREE;30
PH-26-0415;Gériatrie;Enoxaparine 4000 UI;ANNULEE;14
PH-26-0416;Chirurgie;Céfazoline 2 g;PREPAREE;12

manifests/10-dispensing.yaml

apiVersion: connectors.lumnik.io/v1kind: Connectormetadata:  name: scenario-aioff-dispensing-a  connector_type: csv-file  scopes: [pharmacy-a]spec:  transport:    kind: local    path: /tmp/lumnik-scenario/ai-switched-off/a/dispensing-*.csv   # one file per day from the dispensing system  parser:    delimiter: ";"    encoding: UTF-8    has_header: true  after_process: none

$ lm apply -f manifests/10-dispensing.yaml
Applied connector "scenario-aioff-dispensing-a" — 1 endpoint

$ lm connector run scenario-aioff-dispensing-a
run scheduled: c485dde0-c283-4f18-b718-4ca98e597834

$ lm run get c485dde0-c283-4f18-b718-4ca98e597834
ID        CONNECTOR                    ENDPOINT  STATUS     IN  OUT  SKIPPED  DEAD-LETTERS  STARTED           ENDED             EVALUATED
c485dde0  scenario-aioff-dispensing-a  default   Completed  6   6    0        0             2026-11-08 14:13  2026-11-08 14:13  yes

$ lm module list
NAME       VERSION  SCHEMA              STATUS  REASON
core       0.1.0    Platform            ACTIVE  
identity   0.1.0    Identity & Tenancy  ACTIVE  
llm        0.1.0    Model provider      ACTIVE  
rag        0.1.0    Platform RAG        ACTIVE  
search_pg  0.1.0    PostgreSQL search   ACTIVE  

$ lm rag status
indexed: true
chunks: 9
indexed_at: 2026-11-08T13:12:45.463345716Z
provider: ollama-nomic
embedding_model: nomic-embed-text
stored_chunks: 27

$ lm module list
NAME       VERSION  SCHEMA              STATUS    REASON
core       0.1.0    Platform            ACTIVE    
identity   0.1.0    Identity & Tenancy  ACTIVE    
llm        0.1.0    Model provider      INACTIVE  disabled:lumnik.ai.enabled
rag        0.1.0    Platform RAG        INACTIVE  disabled:lumnik.ai.enabled
search_pg  0.1.0    PostgreSQL search   ACTIVE    

$ lm ask --scope pharmacy-a "combien de commandes délivrées hier ?"
Error: this hub has the AI switched off (LUMNIK_AI_ENABLED=false): lm module list shows rag INACTIVE; the operator turns it back on

$ lm rag search "noradrénaline"
Error: this hub has the AI switched off (LUMNIK_AI_ENABLED=false): lm module list shows rag INACTIVE; the operator turns it back on

$ lm scopes
Error: this hub has the AI switched off (LUMNIK_AI_ENABLED=false): lm module list shows rag INACTIVE; the operator turns it back on

$ lm rag status
ai: switched off (LUMNIK_AI_ENABLED=false)
indexed: false
chunks: 0
indexed_at: -
provider: none
embedding_model: none
stored_chunks: 27

dispensing-20261108.csv

order_no;ward;item;status;qty
PH-26-0417;Cardiologie;Furosémide 40 mg;SAISIE;28
PH-26-0418;Réanimation;Propofol 1 %;PREPAREE;10
PH-26-0413;Réanimation;Noradrénaline 8 mg/4 ml;DELIVREE;20

manifests/20-dispensing-lifecycle.yaml

apiVersion: apps.lumnik.io/v1kind: Workflowmetadata:  name: scenario-aioff-dispensing-aspec:  scope: pharmacy-a  provenance: declared  on:    table: connector.t_scenario_aioff_dispensing_a    column: status  states:    - { code: SAISIE, label: "Saisie par le service", initial: true }    - { code: PREPAREE, label: "Préparée par la pharmacie" }    - { code: DELIVREE, label: "Délivrée au service", terminal: true }    - { code: ANNULEE, label: "Annulée", terminal: true }  transitions:    - { from: [SAISIE], to: PREPAREE }    - { from: [PREPAREE], to: DELIVREE }    - { from: [SAISIE, PREPAREE], to: ANNULEE }

$ lm connector run scenario-aioff-dispensing-a
run scheduled: 7581fac3-b112-4484-8ba9-0bc941f57178

$ lm run get 7581fac3-b112-4484-8ba9-0bc941f57178
ID        CONNECTOR                    ENDPOINT  STATUS     IN  OUT  SKIPPED  DEAD-LETTERS  STARTED           ENDED             EVALUATED
7581fac3  scenario-aioff-dispensing-a  default   Completed  3   3    0        0             2026-11-08 14:13  2026-11-08 14:13  yes

$ lm workflow apply -f manifests/20-dispensing-lifecycle.yaml
Applied workflow "scenario-aioff-dispensing-a" → connector.t_scenario_aioff_dispensing_a.status, scope pharmacy-a
  the data matches the declaration
warning: not indexed for chat: the AI is switched off (LUMNIK_AI_ENABLED=false); apply it again once it is on

$ lm module list
NAME       VERSION  SCHEMA              STATUS  REASON
core       0.1.0    Platform            ACTIVE  
identity   0.1.0    Identity & Tenancy  ACTIVE  
llm        0.1.0    Model provider      ACTIVE  
rag        0.1.0    Platform RAG        ACTIVE  
search_pg  0.1.0    PostgreSQL search   ACTIVE  

$ lm workflow apply -f manifests/20-dispensing-lifecycle.yaml
Applied workflow "scenario-aioff-dispensing-a" → connector.t_scenario_aioff_dispensing_a.status, scope pharmacy-a
  the data matches the declaration

$ lm rag search "cycle de vie des commandes de la pharmacie, préparée, délivrée"
SCORE  SOURCE                                SNIPPET
0.698  workflow/scenario-aioff-dispensing-a  Workflow scenario-aioff-dispensing-a — cycle de vie de la colonne status de la t…
0.667  workflow/order-line-cycle-f           Workflow order-line-cycle-f — cycle de vie de la colonne stage de la table conne…
0.656  workflow/acme-order-lifecycle         Workflow acme-order-lifecycle — cycle de vie de la colonne status de la table co…
0.652  workflow/order-lifecycle              Workflow order-lifecycle — cycle de vie de la colonne status de la table connect…
0.647  workflow/scenario-order-sleep-d       Workflow scenario-order-sleep-d — cycle de vie de la colonne status de la table …

Limits

The setting applies to the whole hub: the AI cannot be switched off for one scope only.
The hub must be restarted, once to switch off, once to switch back on.
During the audit nothing is indexed for chat: not a Workflow, not a connector described again, not a file loaded by lm csv ingest. They must be applied again.
lm rag reindex rebuilds only the platform corpus, not the scopes'.
A search by a user without a scope reads every lifecycle of the tenant, not only the pharmacy's.
Orders, wards and quantities synthetic.

Hospital pharmacy · audit

The DPO asks for the AI to be switched off.

For the length of an audit. Without changing edition, without losing data.

LUMNIK_AI_ENABLED=false

The problem

Switched off looked like absent.

rag status said chunks: 0, provider: none.
A lifecycle declared during the audit stayed out of chat, without a word.

What is not changed

The same image, the same data.

The jars stay, the rag schema stays migrated.
The day's files are read as before.

What had to be written

LUMNIK_AI_ENABLED=false

One setting, one restart.

Then true when the audit ends.
And the lifecycle applied a second time.

What was fixed

rag status says ai: switched off and what it keeps (#1895).

The workflow apply warns it does not index (#1896).
One sentence for ask, search and chat (#1894).

Try it

The scenario replays locally.

Every step is published with its output.

This scenario needs lumnik Pro.

See the editions →