camel cmd chaos-explore
Let there be chaos: an AI explores by itself how a running integration copes with faults, for a time budget, and reports what held and what broke
Options
| Option | Description | Default | Type |
|---|---|---|---|
| API key. Also reads ANTHROPIC_API_KEY, OPENAI_API_KEY, WATSONX_APIKEY, or LLM_API_KEY env vars | String | |
| API type: 'ollama', 'openai', 'anthropic', or 'watsonx' | ApiType | |
| How long the AI explores, between 10s and 2h | 10m | String |
| What to explore, such as a route id (default: all routes) | String | |
| Maximum number of tool-calling rounds | 100 | int |
| Model to use. Default: the provider’s default; with Ollama the installed model best at tool calling | String | |
| Save the report in test/chaos/reports (as markdown, and the journal as JSON), and each experiment that failed as a scenario in test/chaos, to replay it | true | boolean |
| Show the tool calls and results as they happen | boolean | |
| Timeout in seconds for one answer of the LLM (an experiment can take minutes) | 300 | int |
| The tools the model gets: core (a small set, for a local model), full, or auto | auto | ToolSet |
| LLM API endpoint URL, as for camel ask. Auto-detected if not specified | String | |
| Print debug information of the LLM calls | boolean | |
| Let the AI send messages to endpoints the integration does not have, and run the other calls that camel ask asks for first, without asking | boolean | |
| Display the help and sub-commands | boolean |