Guides

Bring your own model

Every Kelvar agent runs on a model you choose. Use a hosted provider with your own key, sign in with ChatGPT, or run a model on your own hardware. Kelvar does not resell model access in the Community tier, so your provider bills you directly.

#Providers

Setup lists every provider that is available in your installation. You can connect more than one and choose per agent.

#Local and custom models

Any server that speaks the OpenAI chat API works. Kelvar has no special case for a particular local runtime. You describe an endpoint, and the same provider path serves all of them.

1. Start your server. For example, it might listen on http://localhost:11434/v1.
2. In setup, or under Drivers, add an OpenAI-compatible provider.
3. Enter the base URL and, if required, a key.
4. Enter the model name exactly as the server reports it.

Source code never leaves your machine when the endpoint is local. Read the next section about what the model sees.

#What the model sees

An agent sends the model the task, the files and tool results it reads, and its instructions. That means the provider you choose receives parts of your source code during an audit. Choose a provider whose terms fit your data policy. For code that must not leave your network, use a local model or a gateway you control.

#Models and reasoning level per agent

When you create or edit an agent, you pick a model and a reasoning level. Defaults for the built-in agents are tuned for their jobs. Raise the level for hard audits, and lower it for routine triage to reduce cost and time.

#Cost with your own key

With your own key you pay your provider for tokens. Kelvar's per-run rates in the Usage tier cover orchestration, the sandbox and the ledger, and they do not include model tokens when you bring your own key. Community use on public repositories is free. See Pricing and billing.

#Troubleshooting