Agent releases and configuration

This page covers three related screens under Fleet: Agent Releases, Agent Config, and Cohorts.

Agent releases

Fleet → Agent Releases (/gravity/fleet/releases) is a catalog of published fleet agent binary versions - it doesn’t push anything to your fleet by itself, it just tracks what’s available and how far a version has been rolled out.

Each release has:

• A channel (stable, beta, canary)
• A SHA-256 checksum, used to verify the download’s integrity
• Release notes
• A rollout percentage, tracked in stages (10% → 50% → 100%), which is promoted stage by stage or marked stable to jump straight to 100%

Note

Publishing a new release itself is a Sencai-managed operation, not a self-service action - this catalog is where you can see what’s available and its rollout stage, and it’s what the version field in Force Update (below) draws from.

Rolling a version out to your fleet

Publishing a release here doesn’t move any agent by itself. To actually push a version:

1. Go to Fleet → Agents.
2. Select one or more agents.
3. Click Force Update and enter the target version.

This dispatches an update to every currently connected selected agent - a disconnected agent picks it up the next time it reconnects. The dialog previews which ring (10 / 50 / 100, derived from each agent’s own ID) each selected agent falls into, as a guide for staging your own rollout rather than a separate control.

Each agent independently verifies the downloaded binary’s checksum before replacing itself, and rolls back to its previous version automatically if the new one fails its own health check after being applied - so a bad build can’t silently strand an agent.

Pinning or rolling back a version

There’s no separate “pin” toggle. To pin a host to a version, force-update it to that version and simply don’t push a newer one to it later. To roll back, force-update the affected agents to the last known-good version the same way you’d push any other update.

Agent config

Fleet → Agent Config (/gravity/fleet/config) stores configuration entries - for example, an HTTP proxy for agents running in a restricted or air-gapped network - scoped to your whole organization or to a specific agent, with a Push to agents action.

Caution

Agent-side proxy support is not complete. If you’re deploying into a network that requires a proxy, verify behavior with a test agent before relying on configuration saved here.

Fleet cohorts

Fleet → Cohorts (/gravity/fleet/cohorts) groups hosts by platform and an optional query, on a schedule - a different concept from the version-rollout “rings” described above, despite the similar name.

Note

Deploying a cohort here records it as active but doesn’t currently execute anything against your fleet. For scheduled, fleet-wide checks, use Custom probes instead - they run on every enrolled agent in your organization on a schedule you define.

What’s next

• Fleet agent - what a fleet agent reports once it’s connected
• Agent policies - custom probes, secret injections, and autonomy policies
• Enrollment - installing and connecting a new agent
• Remote execution - running commands and runbooks against your fleet