Connect a provider
Max Harness does not come with model access. You connect the providers you have accounts with, and each request is billed by that provider.
Add a provider from a preset
- Open Providers in the sidebar and press Add provider.
- Pick a provider. The preset fills in the API address and request format.
- Paste your API key. The link under the field opens the page in the provider's console where keys are created.
- Press Connect and test.
The test loads the provider's model list. If that list is public, or the provider has no list endpoint, a one-word message is also sent to confirm that the key works. That message costs a few tokens.
A green Connected badge means you can start chatting. A red Failed badge comes with the provider's own error message.
API keys are not subscriptions
A ChatGPT, Claude or Gemini app subscription does not include API access. API keys come from the developer console of each provider and are billed by usage:
- OpenAI: platform.openai.com
- Anthropic: platform.claude.com
- Google Gemini: aistudio.google.com
- OpenRouter: openrouter.ai
A few coding subscriptions do expose an API, and those have presets: Cline (ClinePass), Command Code and OpenCode Zen. Their plan rules still apply. For example, a ClinePass subscription covers model IDs that begin with cline-pass/.
Add a custom provider
Choose Custom provider at the end of the preset list and enter:
- Name. Shown in the model menu.
- API format. OpenAI-compatible covers most services and self-hosted servers. Use Anthropic or Gemini format only for gateways that copy those APIs.
- Base URL. The API root, for example
https://api.example.com/v1. Leave out/chat/completions. - API key. Sent as a bearer token for the OpenAI format, as
x-api-keyfor Anthropic and asx-goog-api-keyfor Gemini. - Extra request header. Optional, for gateways that require one.
Local models with Ollama
The Ollama (local) preset connects to http://localhost:11434/v1. Two things to know:
- The address is resolved from the web server, not from your laptop. Ollama has to run on the same machine as Max Harness or on a host that machine can reach.
- By default the server refuses provider addresses on private networks. An administrator has to tick Allow providers on local or private network addresses under Admin, Settings.
Default provider
The first provider you add becomes the default for new chats. Change it with Make default on another provider. Each provider also has a default model.
Remove a key
Remove key deletes the stored credential and keeps the connection and model list. Delete removes the connection altogether. Existing conversations keep their messages either way.
Common errors
| Message | Usual cause |
|---|---|
| Did not accept the API key | The key is wrong, revoked or belongs to a different product or region. |
| Billing or credit problem | The account has no payment method or has run out of prepaid credit. |
| Could not find that model or endpoint | The base URL is wrong, or the model was retired. Refresh the model list. |
| Rate limiting | Too many requests for the account tier. Wait, or raise the limit with the provider. |
| Points to a private or local address | Local network providers are switched off in the admin settings. |
Last updated