Agent

Create an agent with a runtime, goal, provider, skills, and schedule.

An Agent runs coding or automation tasks in a Kubernetes pod. Industric supplies its credentials and skills, then starts runs on demand or on a schedule.

Create an agent

Dashboard: click Create agent on the Agents or deployment page. Choose an online deployment, enter a lowercase DNS-style name, pick a runtime, and write a goal. OpenCode offers a free, keyless model. Claude Code and Codex need a provider.

The Create agent dialog: name, target deployment, OpenCode runtime with the runtime default provider, MCP servers, skills, tools, goal, and addons — Choose a deployment, runtime, and goal. New agents start Paused.

YAML: define an Agent resource and apply it with kubectl or your GitOps tool.

A sample Agent

agent.yaml· yaml

1apiVersion: autonomic.sh/v1alpha1
2kind: Agent
3metadata:
4  name: triage-bot
5  namespace: adaptive
6  labels:
7    autonomic.sh/agentgroup: support-team
8    env: prod
9spec:
10  runtime: ClaudeCode
11  displayName: Issue triage
12  phase: Active
13  goal:
14    inline: |
15      Triage incoming GitHub issues: fetch new ones hourly, label, assign.
16  provider:
17    name: anthropic-prod
18    type: anthropic
19  gateway:
20    name: shared-gateway
21  wakeup:
22    enabled: true
23    mode: schedule
24    schedule: "0 * * * *"
25  bootstrap:
26    skills:
27      - triage-runbook
28    mcps:
29      - linear
30    policies:
31      - safe-shell-baseline

Fields

runtime: agent software and image. See Runtimes.
displayName: an editable display name. The resource's name cannot change.
labels: narrow apply-to-all command policy coverage. Every configured label pair must match the agent's metadata.
phase: Active or Paused, defaulting to Paused. Check status before setting Active.
goal: inline text or a configMapKeyRef.
model: deprecated and ignored. Set the model in the provider, or use the runtime default without a managed provider.
provider: supplies credentials. Use name for a specific provider or type for that type's tenant default.
agentToken: a Secret reference for a runtime API key when no managed provider is used.

Capabilities

These flags control agent behavior. All default to off, allowing connections and runs:

disableShell: block interactive terminal access.
disableRun: hide Run and block runs from webhooks and schedules.
injectGoals: send the goal directly into the session.
allowUserContext: let users add instructions for one run.

mode is legacy

spec.mode (supervised / system) still is accepted in older manifests. The flags above control behavior, including runs for supervised agents.

Runtimes

Choose ClaudeCode, Codex, Gemini, Pi, Hermes, Hal, or OpenCode. See providers, examples, and image versions in Runtimes.

What you can attach

Add or change attachments in the dashboard or agent resource:

Attachment	Field	What it gives the agent
Skills	`bootstrap.skills`	Reusable instructions loaded at session start.
MCP servers	`bootstrap.mcps`	Remote tools, with URLs and credentials supplied at connection time.
Policies	`bootstrap.policies`	Policy bundle names; delivers their command and MCP members to supported agent enforcement paths.
Memories	`bootstrap.memories`	Saved knowledge agents can read and update.
File systems	`filesystems`	Shared storage that survives restarts.
Gateway	`gateway`	Model access with usage tracking and budgets.
Web agents	`webAgentRefs`	A browser the agent can drive, over MCP.
Schedule	`wakeup`	Fire runs on cron, a phrase, or once at a fixed time.

Resolved at session start

Skills, policies, and MCP servers are loaded by name at session start. Their updates apply on the next run without editing the agent.

Templates

Save an agent template to reuse its runtime, goal, provider, capabilities, and attachments. Choose a built-in or custom template in the dashboard, CLI, or API, then override what you need. Templates belong to the tenant and do not include an agent name.

Pause, restart, delete

Pause sets spec.phase: Paused, removes the resident pod, and stops schedules. Resume to bring it back.
Restart replaces the resident pod. Active runs use separate pods and continue.
Delete removes the agent, pod, run configuration, and secrets. Force delete clears its dashboard record if the deployment is offline or unresponsive.

Status

Run kubectl get agents to check .status.phase and .status.ready. For details or a run that won't start, use kubectl describe agent <name>.