Tools
Crush
Set up Crush with Cognitivers, then verify your connection.
Connects directly
Charm's terminal agent. The provider type matters: openai-compat, not openai.
1. Before you begin
- Crush installed and ready to configure.
- A Cognitivers API key. See authentication and key management.
Configuration location
crushrc in the project or ~/.config/crush/crushrc
Need an API key? Request access before running these examples.
2. Configure the connection
Apply the following configuration in the location above. Replace credential placeholders with your own key; keep secrets out of version control.
# Add the provider.
provider add cognitivers \
--type openai-compat \
--base-url "https://api.cognitivers.com/v1" \
--api-key "${COGNITIVERS_API_KEY:?set COGNITIVERS_API_KEY}"
# Register the models.
model add cognitivers/cog-pro --name "Cog Pro" --context-window 262144
model add cognitivers/cog-fast --name "Cog Fast" --context-window 262144Compatibility notes
- Use openai-compat. The openai type is for routing through OpenAI itself.
- Configuration moved to a Bash crushrc file. crush.json still works but is deprecated and receives no new options.
- Both files run as code: keep them out of untrusted hands.
3. Verify the setup
Use this prompt in your configured tool to check authentication, a response and streaming. It reports failures without printing your key.
Open verification prompt
You are working in a repository where I have configured Crush to use Cognitivers as its model provider.
The API is OpenAI-compatible chat completions.
- Base URL: https://api.cognitivers.com/v1
- Streaming: use the same base URL with stream: true
- Models: cog-fast (agents, pipelines, volume), cog-pro (one hard task), cog-embed (retrieval)
- Authentication: an API key in the Authorization header as "Bearer sk-cog-..."
Do these four things and then stop.
1. Check that the credential is present in my environment and tell me the name of the variable you found it in. Do not print the key itself, not even partially.
2. Send one non-streaming chat completion asking for a single short sentence, using cog-fast. Show me the exact command or code you used and the model's reply.
3. Send the same request with streaming enabled and confirm you received server-sent events, not one buffered body.
4. If any step fails, tell me the exact error, the status code and the request that produced it, then stop. Do not try more than two fixes, and do not change any configuration file without showing me the diff first.
Rules: never write the key into a file that is tracked by git, never print it in full, and if the request works, say so in one line instead of writing a report.
5. Confirm the provider type is openai-compat and not openai, and tell me whether the model discovery ran against the API or whether you registered the models by hand.If the check fails, use the reported status code in the error reference and review the compatibility notes above.