One workspace, many models
Models are a replaceable capability in LoneAct, not the product itself. A workspace chooses a default model from the catalog, the Composer lets you switch per conversation, and each model can have its own context window, tool support, pricing, and provider routing.
Pick by task
Use a stronger model for complex planning and code changes, and a faster, cheaper model for lightweight organization work.
Workspace default
Set the default model for new conversations in Settings → Models so you do not choose every time.
Role-level override
An Agent role can specify its own model; when it does not, it inherits the workspace default.
Use your own model key
If a model supports your own key, you can configure a route in the model selector. The key is used only by LoneAct on your computer when requesting the model service; it is not written into the conversation and is never sent to the model as part of the prompt.
Open the model selector
Choose the target model in the Composer model menu and look for the own-key marker.
Configure provider protocol and endpoint
Fill in the provider protocol, model name, and Base URL supported by the model, then enter the API key.
Save and verify
After saving, the model shows as configured and the next request uses your route. You can edit or replace it when needed.
Your keys are managed by LoneAct secure storage. Use a dedicated key with the smallest scope, and do not use a production master key for experiments.
How to choose: by task, not by leaderboard
| Task | Recommendation | Why |
|---|---|---|
| Cross-file feature work, complex refactoring | Prefer a model with stable tool calling and a longer context | It must read files, edit, and run verification continuously |
| Understanding a codebase, shaping a plan | Choose a model with stronger reasoning and long-context ability | It must turn scattered module relationships into structure |
| Batch cleanup, simple queries | Choose a faster, lower-cost model | The task boundary is clear and requires less complex reasoning |
| Sensitive projects | Choose a provider and deployment that meet your organization's compliance requirements | The model service boundary must match your data policy |
Common questions
Can I switch models mid-conversation?
Yes. The Composer model choice applies to the current conversation; the workspace default only affects new conversations.
Can different Agents use different models?
Yes. Roles support model overrides, so planning, execution, and review roles can use different models.
What if a model is unavailable?
The model selector shows the reason; switch to an available model or check the provider connection and retry.
Am I locked to one provider?
No. LoneAct's model catalog and routing layer support multiple providers; the exact list depends on your workspace configuration.