Pool operations
Once your pool is deployed and active, you can manage it from the Private Pools page in the portal.
To keep its Operator, Armor and other images patched and on the latest release, see Keep your pool's images up to date.
Assigning MCP Servers to a Pool
- Navigate to your MCP server in the portal
- Select Deploy to Private Cloud
- Choose the target pool and enter an MCP prefix (for example,
crm-api) - The Operator picks up the change and deploys the MCP server — typically within 60 seconds
- Your MCP server is now accessible at
https://{pool-host}/{mcp-prefix}/mcp
Force Sync
If you've made configuration changes and want them applied immediately (rather than waiting for the next sync cycle), use the Force Sync action from the pool detail page. This tells the Operator to re-read its configuration and reconcile immediately.
Reseed
If a pool's configuration becomes out of sync (shown in the pool health dashboard), use the Reseed action to rebuild all configuration from the source of truth. This is a safe operation — it regenerates configuration without affecting running traffic.
Monitoring Pool Health
The Private Pools page shows the health of all pools at a glance:
- Active pools with recent heartbeats are healthy
- Unknown pools (no heartbeat in 30+ minutes) may have networking issues
- Out of sync pools have configuration drift and may need a force-sync or reseed
For detail beyond these states, such as request rates, latency or how much sensitive data the DLP engine finds, collect the pool's metrics in your own monitoring tools. See Monitoring and metrics.
Isolation Tiers
A single pool often hosts workloads with very different risk profiles — a production server you depend on sitting next to one you are still evaluating. Isolation tiers let you give your most important apps, servers and LLMs their own Armor with dedicated CPU and memory, so work in one tier never competes for resources with another, and a misbehaving workload cannot disrupt your production traffic.
By default a pool runs one Armor for all of its traffic. Turn tiers on when you want that separation.
Isolation tiers are available on private pools only. Cequence Cloud pools always run as a single Armor. Turning tiers on or off, renaming them and changing their resources requires tenant administrator access.
The built-in tiers
| Tier | Intended for |
|---|---|
| Tier 1 | Highest priority. Business-critical apps and LLMs that need dedicated resources and your fastest compute. |
| Tier 2 | Everyday workloads on shared resources. This is the default tier. |
| Tier 3 | Lower priority. New, experimental or batch workloads that can run on lighter resources. |
Pools that enabled isolation before tiers could be named show these as Critical, Standard and Experimental instead. They behave identically.
You can rename any tier, and add your own — up to 8 tiers in total per pool.
Turning tiers on
- Open the pool, go to the Configuration tab and select Isolation Tiers.
- Select Turn on.
- Choose at least two tiers, and which one is the default.
- Save.
Each enabled tier gets its own Armor, and a tier router takes over the pool's traffic. The current Armor keeps serving until the new ones are ready.
Two tiers is the minimum — a router with only one tier has nothing to route between.
Adding and sizing a tier
From the Isolation Tiers section, Add tier creates your own alongside the built-in ones. Each tier has:
- a Name of up to 40 characters, which you can change at any time
- a Short name, generated from the name and fixed once saved
- an optional description of what belongs in it, shown wherever someone picks a tier
- Resources — either the pool's Armor defaults, or custom CPU, memory and replica settings
Give higher-priority tiers more CPU, memory and replicas. This is what "dedicated capacity" means in practice: each tier is a separate Armor deployment, scaled on its own.
Every enabled tier runs its own Armor deployment in your cluster. Enabling three tiers means three Armors plus a router, so make sure the cluster has capacity before you turn them on.
Adding custom tiers requires an upgraded data plane. If the pool's operator and Armor do not support it yet, Add tier is disabled with a note telling you to upgrade. The three built-in tiers work regardless.
Assigning a server, persona or LLM to a tier
When you add an MCP server, persona or LLM to a pool with tiers — or open its configure dialog afterwards — you will see an Isolation Tier selector. The pool's MCP servers table also shows each server's tier in an Isolation Tier column.
- Anything you do not assign runs on the pool's default tier, which is Tier 2 unless you changed it. New resources are not put in a particular tier automatically.
- The selector only offers tiers the pool actually runs, so nothing can be pinned to a tier that does not exist.
- Assigning to Tier 1 is restricted to administrators.
Automating with the API?
cellis accepted on the MCP-server and persona create and update endpoints, andcellson the pool endpoints, so scripted workflows can assign tiers directly.
Changing the default, or turning a tier off
The row menu on each tier offers Edit, Make default and Turn off (or Remove, for a custom tier).
- Make default changes where unassigned resources run.
- Turning off a built-in tier keeps its sizing for when you turn it back on. Removing a custom tier discards its name and sizing.
- Either way, anything assigned to that tier moves to the default tier and keeps serving — its Armor is only removed once the default tier is ready.
You cannot go below two tiers; to get back to a single Armor, use Turn off on the Isolation Tiers section itself.
Tips
- Check pool health regularly from the Private Cloud page
- Watch for stale heartbeats — they may indicate networking issues between your cluster and the control plane
- Use force-sync after configuration changes for immediate application
- Use
aigateway monitorfor comprehensive real-time monitoring