faramesh.devBETA
docsgithubcommunityget started →

docs

Getting Started

Core

FPL

Governance

Operations

Reference

github ↗slack ↗community ↗

16 , fleet

Fleet Operations

Fleet operations are where policy quality meets operational reality. A single-agent setup can tolerate manual guardrails; a fleet cannot. As soon as you run many autonomous agents across services, regions, or teams, you need deterministic rollout, fast containment, and clear blast-radius control for policy and runtime changes.

Faramesh fleet commands are designed for two goals: consistency and containment. Consistency means policy state does not fragment across instances. Containment means when one segment misbehaves, you can isolate quickly without freezing unrelated workloads. This is the difference between a targeted incident response and a platform-wide outage.

Fleet rollout model

Rollouts should be progressive and observable. Start narrow, verify decisions and error rates, then widen scope.

01Canary: deploy policy updates to a small subset of agents first.
02Observe: monitor deny/defer spikes, approval queues, and latency impact.
03Expand: widen rollout after behavior is stable and expected.
04Contain: trigger kill only for affected targets when risk crosses thresholds.

List fleet agents

List registered agents and current health/trust status.

bash
faramesh fleet list

Push policy

Push a policy update across a fleet scope. Include a rollout message so operators can map behavior changes back to a specific control event during audits and incident triage.

bash
faramesh fleet push <instance-id> --message "policy rollout"

Fleet kill switch

Emergency deny for all tool execution across selected fleet targets. Use kill for containment, then restore by policy correction and staged re-enable, not by blindly unfreezing everything.

bash
faramesh fleet kill <instance-id> --reason "incident containment"

Fleet anti-patterns

These mistakes cause the most avoidable outages in large deployments:

!Pushing global policy changes without a canary cohort.
!Using kill switches as routine maintenance instead of targeted containment.
!Ignoring gradual deny/defer trend drift until it becomes customer-visible failure.
!Running inconsistent policy versions across regions with no reconciliation cadence.

At fleet scale, governance failures are distribution failures as often as policy logic failures. Rollout discipline is a security control.

Need help? Use any of these support options.

open an issuejoin slackcommunity/forum
MIT License
docsgithubcommunityget started →