ark-model-configuration.md

Ark Model Configuration Notes

This note records the model configuration strategy used during the OpenClaw cron repair.

Goal

Scheduled jobs should consistently use the Volcengine Ark model stack unless a job has a documented reason to use something else.

Desired shape

Global defaults:

• Primary model: volcengine/ark-code-latest.
• Fallback models: Ark-compatible fallbacks.
• Cron job model overrides: 0.

Cron jobs should inherit global defaults instead of storing their own model value.

Why remove per-job overrides

Per-job model overrides create silent drift:

• Old jobs keep using deprecated providers.
• Disabled jobs can become broken when re-enabled later.
• Debugging becomes harder because scheduler behavior differs per task.
• Migration to a new provider is incomplete unless every job is audited.

Removing overrides makes the actual model path easier to reason about.

Verification commands

List cron jobs with explicit model overrides:

openclaw cron list --all --json \
  | jq '[.jobs[] | select(.payload.model != null) | {name, model: .payload.model}]'

Expected result:

[]

Run an explicit gateway smoke test:

openclaw infer model run --gateway --model volcengine/ark-code-latest

Expected behavior:

• The model call succeeds.
• The provider shown by logs/status is Volcengine Ark.
• The command returns a known marker if using a marker prompt.

Expired non-Ark auth

Expired OpenAI or Qwen auth profiles do not block Ark-based cron execution. Reauthenticate them only if they are still expected to be used:

openclaw models auth login --provider openai

For legacy Qwen portal profiles, follow the current OpenClaw reauth flow printed by openclaw doctor.

Migration checklist

• Confirm the global default model is Ark.
• Confirm Ark fallback models are configured.
• Remove cron-level payload.model values.
• Verify enabled and disabled jobs both have no overrides.
• Run at least one Ark smoke test.
• Let write-path jobs validate naturally or run one narrowly scoped manual test.