Help Center / Manage Access, Billing, and Limits
ReferenceAgentic Workers Supervisor FAQ
See how many agents can run on a device, who pays, and what happens when it goes offline.
On This PageHow Many Supervisors Can I Install on One Device?
- How Many Supervisors Can I Install on One Device?
- How Many Agents Can Run on One Device?
- Can I Connect Several Devices?
- Will It Find My Existing Hermes Agents?
- Who Pays for Models and Compute?
- Why Are Codex Models Missing if ChatGPT Is Connected?
- What Happens When the Device Goes Offline?
- Can I Still Use Telegram and Local Tools?
- What Happens if I Disconnect, Uninstall, or Lose Ultra?
How Many Supervisors Can I Install on One Device?
Use one Supervisor for the device’s connected environment. It manages multiple agents. On macOS, the managed service uses one system-wide service identity, so a second install cannot take over that service. On Linux, the managed service is scoped to the local account, so the installer permits one managed Supervisor per account.
Separate Linux accounts have separate user services, but this is not a tested multi-workspace device deployment. Use separate devices for separate client environments. Changing the install folder does not create another independent service for the same account.
One environment is bound to one workspace. Do not start two Supervisors with copied credentials; only one process can own that environment’s active connection.
How Many Agents Can Run on One Device?
Multiple agents can run on one device, with one selected Hermes profile per agent. The current default is 10 Super Agents across the workspace, shared between cloud and local agents. A workspace with 3 cloud Super Agents normally has room for 7 more connected Super Agents, not 10 more for each computer.
A workspace can have a configured allowance that differs from the default. The creation or pairing flow checks the actual allowance. Adding an empty environment does not create an agent or use an agent slot.
The environment can report at most 64 profiles. That is a technical inventory limit, not a promise that you can create 64 agents or run 64 tasks at once. Available memory, CPU, provider limits, and workspace capacity still apply.
Can I Connect Several Devices?
Yes. A workspace can have several environments. Choose a connected environment when creating an agent. There is no separate numeric environment allowance enforced by the current enrollment flow; the shared Super Agent allowance still limits the agents you add.
An agent stays on its selected environment. This release does not move an existing cloud agent onto a device, provide remote desktop control, or back up and restore device data in the cloud.
Will It Find My Existing Hermes Agents?
Setup detects profiles under the chosen local account’s Hermes locations and lets the owner select them. It does not scan every account or every folder on the computer. Unselected profiles are not added to the workspace.
Each selected profile retains its own memory, sessions, skills, credentials, and approval settings. A shared or ambiguous running gateway must be resolved before the Supervisor can manage that profile safely. Do not run a second gateway or scheduler for it.
Profile discovery happens during setup. It is not a continuous scan that automatically imports new local profiles. Create additional managed agents from Agentic Workers; ask support before trying to add another existing profile to an already connected environment.
Who Pays for Models and Compute?
Choose Agentic Workers credits or credentials on the device for each profile. Existing profiles default to their current credentials. Fresh profiles default to Agentic Workers credits.
AGW-funded model usage is charged to the workspace’s billing owner through the existing credit system. Local-provider inference has no platform inference debit; the provider’s own charges and limits still apply. A connected device incurs no Fly compute charge. You supply the device, power, and network.
A connected ChatGPT account can provide eligible Codex models when that agent’s Hermes profile reports a usable connection. ChatGPT plan limits still apply. Connecting ChatGPT does not make other provider models free or switch every run away from AGW credits.
Why Are Codex Models Missing if ChatGPT Is Connected?
Native ChatGPT connection requires Supervisor connector 0.2.8 or later and compatible Hermes authentication support. “Device setup needs attention” means the connector has not advertised that support. Choose Fix device setup for diagnostics and the update guide. On the agent’s computer, check agw-device status and agw-device doctor. Update only to a published compatible release; if already current, resolve the Hermes diagnostics and restart the connector before choosing Check again. A saved account is not always usable. Open that agent’s Configure, Provider & model, then ChatGPT account. “Connection needs attention” means Hermes cannot use the saved account. Choose Reconnect ChatGPT and finish sign-in, then check the model selector again.
“Could not verify ChatGPT connection” means verification was unavailable. Retry after checking the device connection. The model list shows only models supported and authenticated for that selected profile; it does not list every provider advertised elsewhere in the app.
What Happens When the Device Goes Offline?
New work is rejected while the device is offline. Keep your draft and retry after reconnecting. Work is not silently queued or moved to a cloud agent.
An existing run can reattach with its original run ID if Hermes can recover it. A process restart may interrupt it. Do not resend the same task just because the connection dropped; first check whether it completed or was interrupted.
Routines stay scheduled in Hermes on the device. Completed results are buffered for deduplicated delivery after reconnecting. The Supervisor does not add a new policy to replay missed schedules.
Can I Still Use Telegram and Local Tools?
An existing Hermes Telegram gateway can keep using the selected profile’s channel settings under Supervisor management. Send a message to that configured bot and test its reply. Only one gateway should poll the same bot token. The Agentic Workers Telegram integration is a separate channel flow.
The local account’s file and command access applies. Workspace member chat and owner management permissions still apply in Agentic Workers. A member’s ability to ask an agent to run a local tool is not an operating-system sandbox.
Enable integrations for the selected agent and test them. Local provider credentials are not uploaded. The device receives scoped access, not the workspace’s entire collection of secrets.
What Happens if I Disconnect, Uninstall, or Lose Ultra?
Disconnecting the environment ends cloud access for its managed agents. Disconnecting one agent ends that agent’s binding. Revocation does not delete local memory, skills, sessions, or workspace files.
Uninstall removes the managed service and command while preserving agent data and recovery files. Setup-changed settings and prior services are restored where they have not since been edited. Conflicting later edits are preserved for manual review.
Losing Ultra or the enrolling owner’s authority blocks platform access without deleting local agent data. If revocation cannot be confirmed during an offline uninstall, the owner must disconnect the environment in workspace settings.
Was this helpful?
Your feedback helps us rank and improve Help Center articles.
Related Articles
Install Agentic Workers Supervisor
Connect a client device, choose existing Hermes agents, and create more agents in the same environment.
Read ArticleTroubleshoot Agentic Workers Supervisor
Fix setup errors, missing agents, offline devices, and model sign-in problems.
Read ArticleManage Credits and Plan Limits
Understand how credits, usage, and plan limits affect agent runs.
Read Article