cron-model-fallback

name	cron-model-fallback
description	Cron job model fallback chain for OpenClaw. User-configurable model priority chain — tries each in order until one succeeds. No provider-specific logic.
homepage	https://looi.ru/a/looi-clawd
metadata	{"author":"Arthur Arsyonov","license":"MIT"}

cron-model-fallback

User-configurable model fallback chain for OpenClaw cron jobs.

Your cron job says "use model X". Model X is down. Job fails silently. This script fixes that — configure a priority chain, it tries each model until one works.

Problem

OpenClaw cron jobs accept a single model. If that model is unreachable, the job fails silently — no retry, no fallback, just lastRunStatus: "error".

Solution

A zero-dependency wrapper that:

1. Takes a list of models in priority order (CLI flag or config file)
2. Calls openclaw agent --model <model> for each, in order
3. Returns the first successful response
4. No provider-specific logic — works with any model OpenClaw supports

Usage

# Explicit chain
python3 fallback.py \
  --models "anthropic/claude-sonnet-4,google/gemini-2.5-flash,groq/llama-3.3-70b" \
  --prompt "Summarize today's news"

# Defaults from config (see Configuration below)
python3 fallback.py --prompt "Your task"

# With a specific agent
python3 fallback.py \
  --models "google/gemini-2.5-flash,anthropic/claude-haiku-4-5" \
  --agent my-agent --prompt "Your task"

Configuration

Default fallback chain (optional)

Add to ~/.openclaw/openclaw.json:

{
  "cron": {
    "fallbackModels": [
      "anthropic/claude-sonnet-4",
      "google/gemini-2.5-flash",
      "groq/llama-3.3-70b-versatile"
    ]
  }
}

When --models is omitted, the script reads this array. This means every cron job inherits the same fallback chain without per-job configuration.

Authentication

Token is read from (in priority order):

1. OPENCLAW_TOKEN environment variable
2. ~/.openclaw/openclaw.json → gateway.auth.token

Parameters

Parameter	Default	Description
--models	from config	Comma-separated model list, first = most preferred
--prompt	—	Task prompt string
--prompt-file	—	Path to file containing the prompt
--agent	(default)	OpenClaw agent name (optional)
--timeout	120	Seconds to wait per model attempt
--max-tokens	4096	Maximum tokens in the response
--quiet	false	Suppress progress output

Architecture

fallback.py
  │
  ├─ Resolve model chain: --models flag → cron.fallbackModels config
  │
  └─ For each model:
       └─ openclaw agent --model {model} --message {prompt} --json
          ├─ Success → print result, exit 0
          └─ Failure → try next model

No provider-specific code. No health checks. No API parsing. Just: call the CLI, check the exit code.

Exit Codes

Code	Meaning
0	Success
1	All models failed
2	Configuration error (no models, no token, no prompt)

Requirements

• Python 3.8+ (stdlib only, no pip dependencies)
• OpenClaw CLI in PATH