Choose models for the AI agent
The agent works through tools, so the model it uses must support tool use. By default the platform chooses a suitable model for you.
Auto (default)
Section titled “Auto (default)”With Auto, CHAMPREP routes each turn, and each subagent task, to a model
that fits the request and supports what the turn needs, chosen from the models
available to you. It is the recommended choice, and the status line shows
Auto.
Return to Auto at any time with /model auto or by choosing
Auto (recommended) in /models.
Pick a model
Section titled “Pick a model”/modelsThe picker lists:
- Auto (recommended).
- Platform models, grouped by provider. Each shows capability badges:
toolsorno tools,vision, andlong context. Models marked 🔒 are not in your plan and cannot be selected. - Your models: models you registered with your own provider key.
- Register a new model.
Type to filter, then press Enter. You can also switch directly with
/model NAME or start a session with champrep ai --model NAME, where
NAME is a platform model ID, the name of a model you registered, or auto.
Your choice is remembered for the current project in
.champrep/settings.local.json. Only model identifiers are stored there,
never keys. Subagents use the same selection.
From a script, list the models available to you with:
champrep ai models lsWhen a model can’t do something
Section titled “When a model can’t do something”- No tool support. If you pick a model marked
no tools, the agent warns you first: without tools it cannot read or edit files, run commands, or use CHAMPREP services. You can switch to Auto, keep the model for plain chat, or pick another. - Missing capability. If a turn needs something the chosen model lacks, such as image input or a larger context window, the turn is refused and nothing is changed. The agent offers to switch to Auto and resend the prompt, or to pick another model. It never switches for you.
Use your own model key
Section titled “Use your own model key”You can register a model that uses your own API key from a supported provider. Requests with that model are billed by your provider on your own account and do not use your CHAMPREP AI credits.
Supported providers are Anthropic, OpenAI, Google Gemini, xAI, DeepSeek, and Moonshot Kimi. Registration shows the models available for each provider.
Register in the agent
Section titled “Register in the agent”/model register- Paste your provider API key into the masked field.
- Confirm the provider. It is detected from the key where possible; otherwise you choose it.
- Choose the model and give it a name, for example
My Claude.
The key is checked with the provider before it is saved. Select the new model
in /models, where it is labelled your key, or with
champrep ai --model "My Claude".
Register from the command line
Section titled “Register from the command line”The key is read from a hidden prompt, or from standard input with
--key-stdin. It is never accepted as a flag or environment variable, so it
cannot end up in shell history or process lists.
champrep ai models register --name "My Claude"pass show anthropic | champrep ai models register --name "My Claude" --key-stdinOptional flags: --provider (otherwise detected from the key) and --model
(otherwise the provider’s first tool-capable model).
How your key is handled
Section titled “How your key is handled”- It is entered only in a masked field or from standard input, and is never shown again, printed, or written to the transcript, prompt history, or logs.
- It is sent once to your CHAMPREP AI account and stored there, not on your machine. Only a masked fingerprint is shown afterwards.
- It is used only for your own requests with that model.
- Keys pasted into a normal prompt are blocked and never sent to the model.
Remove a key
Section titled “Remove a key”/model remove "My Claude"champrep ai models rm "My Claude"Removing a model deletes its key from your CHAMPREP AI account. You can also manage your keys in your CHAMPREP AI settings on the web.
Reasoning effort and context
Section titled “Reasoning effort and context”/effort low|medium|high|max|offsets reasoning effort for models that support it./contextshows how much of the model’s context the conversation uses./compact [focus]summarises the conversation to free context, keeping what you name.