start_within per request, so the deadline goes on the key. Create one first (see agent keys).
OpenWorker is in open beta, so these fields move. Check them against your
build.
Configure OpenWorker
- Open OpenWorker and go to provider settings. Pick the OpenAI provider.
- Paste your agent key into OpenAI API key.
-
Set Custom endpoint (optional) to the router, including the
/v1suffix. -
Set the default model in
~/.config/coworker/config.toml.
<your-project>/.coworker/config.toml overrides the global one. Every key in it is optional, and unset keys fall back to built-in defaults.
OpenWorker also reads OPENAI_API_KEY from the environment. Export your agent key there to keep it out of the secret store.
Confirm the key applied
Every response comes withx-flexinference-defaults-applied. OpenWorker doesn’t show response headers, so read the request in the dashboard under Logs instead.
Troubleshooting
OpenWorker didn’t detect the provider from the key. It checks forsk- and AIza prefixes to guess one, and a flex_live_ key matches neither. Pick OpenAI yourself.
Requests still go to OpenAI. Custom endpoint (optional) is blank, so the provider falls back to api.openai.com. Set it to the router URL, including /v1.
400 unsupported_parameter. OpenWorker sent a Chat Completions parameter we don’t translate, and on a non-OpenAI model we refuse any one of them. Use a gpt-* slug.
400 key_default_not_applicable. Your key holds a duration and one of those parameters came beside it. Edit the key to a tier.
The model picker and config.toml disagree. A per-workspace .coworker/config.toml overrides the global file. Check which one your project directory has.
See errors for every refusal we return, and agent keys for the ones that are not specific to OpenWorker.