Feature Toggles

Feature toggles flip optional or experimental behaviors for the whole installation. They live in the FeatureToggle table of the 2501 database, one row per key, and you change them there. Each time it starts, the engine writes a row for every key below that has none, keeps the state of the rows that exist, and deletes the rows of keys it no longer knows. After the first start, the table holds exactly the keys below. Operational rules and host facts are always checked before proposed task actions run when either is available. This verification has no feature toggle.

Open a database session

With the PostgreSQL 2501 deploys on Kubernetes:

kubectl -n "$NS" exec -i deploy/postgres -- psql -U 2501 -d 2501

On Docker Swarm, with $STACK the name you deployed the stack with (ai-2501 in Deploy):

docker exec -i "$(docker ps -q -f "label=com.docker.swarm.service.name=${STACK}_postgres")" psql -U 2501 -d 2501

With your own PostgreSQL, connect with the DIRECT_URL from env.engine:

psql "$DIRECT_URL"

List and change toggles

-- Every toggle and its state
SELECT key, enabled, description FROM "FeatureToggle" ORDER BY key;

-- Turn one on (set false to turn it off)
UPDATE "FeatureToggle" SET enabled = true, updated_at = now() WHERE key = '<key>';

Command Center reads toggles on each request, so a change shows after a page reload. The engine reads them when it starts: restart it to apply a change, with kubectl -n "$NS" rollout restart deploy/engine or docker service update --force "${STACK}_engine". Turning on enableSingleSignOn or enableLogLlmInputAndOutput this way asks for no authenticator code, since database access already grants everything.

Reference

Every toggle is off on a new installation, except two the first run sets from values.yaml: enableSatelliteClusters from features.satellites, and enableElasticSearch when elasticsearch.mode is not none.

Integration

Toggle Key	Description	Default
`enableElasticSearch`	Enable Elasticsearch integration	❌ Disabled
`enableDockerSwarm`	Enable Docker Swarm integration	❌ Disabled
`enableSatelliteClusters`	Register customer Kubernetes clusters and place a host’s executors in one	❌ Disabled
`enableCentral`	Enable Central configurations, agent targets and execution	❌ Disabled
`enableDiscovery`	Enable network discovery (subnet scan, authenticated recon, host promotion)	❌ Disabled
`enablePrivilegedAccess`	Read credentials from a PAM system: the Privileged access page and the PAM account credential type	❌ Disabled
`gateways.allow_ecritel_type`	Enable the Ecritel gateway: its type in the create form, and the polling that opens tickets	❌ Disabled

Security & Logging

Toggle Key	Description	Default
`enableSingleSignOn`	Enable single sign-on: the Security settings section, the SSO button on the login page, and pointing a tenant at an identity provider	❌ Disabled
`enableLogLlmInputAndOutput`	Log LLM input and output	❌ Disabled
`enableLlmCorrelationHeaders`	Send the ticket, job, task and agent IDs of each LLM request in `X-2501-*` headers; read at engine start	❌ Disabled

Jobs

Toggle Key	Description	Default
`enableFollowUpJobs`	Detect and create follow-up jobs after terminal jobs	❌ Disabled
`enableJobRestart`	Restart a running job when the user adds a comment	❌ Disabled

Experimental

Toggle Key	Description	Default
`enableSlowInference`	Reduce parallel tasks/jobs for slow inference	❌ Disabled
`enableCompletionGate`	Send a task back once when it reports done right after an unchecked host change	❌ Disabled
`enableObservability`	Accept monitoring alerts as signals and let Sentinel open Investigations; read at engine start	❌ Disabled

When to flip what

enableElasticSearch if you have an Elasticsearch instance and want the engine + Command Center to ship logs to it.
enableLlmCorrelationHeaders when your LLM proxy groups requests by ticket, job or task, for example to build Langfuse sessions.
enableSlowInference if your tenant runs on a self-hosted model with low throughput - reduces task concurrency to avoid queue starvation.
enableSatelliteClusters to reach hosts from inside a customer Kubernetes cluster instead of dialing them from the engine. It adds the Clusters page and a host’s executor placement fields. Turning it back off hides that surface without changing where anything executes: a host already placed on a cluster keeps running through it, and placement stays writable through the API, so the toggle governs the surface rather than the capability.
- While it is off, two states have no way back through the Command Center: a host cannot be returned to a direct dial, and a credential held by a cluster cannot be deleted. Turn the toggle on to unwind either.
Anything labeled experimental: open a support ticket before enabling in production.

​Open a database session

​List and change toggles

​Reference

​Integration

​Security & Logging

​Jobs

​Experimental

​When to flip what