BRIEFLOOP / GUIDES
Models and costs
Understand execution, data recipients and billing.
Current installers: Mac 0.31.7, Windows 0.31.7. Labels and capabilities vary by channel; the installer version is authoritative. Development features are not implied.
Before you start
Use a workspace you can back up. Model tasks require an authenticated runtime and permission to process the chosen materials.
Steps
- Choose BriefLoop Agent or Agent CLI in settings. Configure a model service directly for the built-in agent; for a CLI, sign in or configure its API in that tool first. Usage is charged to the selected service account.
- Use a model and reasoning option that the current runtime supports. Version 0.25.0 supports OpenCode 1.x and 2.x; configure v2 providers in OpenCode. Saving settings is not a successful connection test: verify with a short public-text conversation.
- The main agent, Evaluator, Wiki Maintainer and Skill Proposer can use separate models. Evaluation and learning may also consume model usage.
- Set research limits, length and execution time. These are not a complete monetary cap; metering of runtime-native search varies.
- Check actual quotas and bills with the model and search providers. BriefLoop does not supply their subscriptions or model credits.
- Jev uses separate TypeSafe credentials and quota, outside your writing-model, OpenCode Go or CLI subscription. It sends the displayed material only when you start a precheck.
What success looks like
The first conversation should return an actual response. New tasks use saved settings; queued and running tasks retain their original configuration.
If something goes wrong
For quota errors, 429, 503 or an unknown model, inspect the task error and runtime authentication. Matching model names do not imply shared subscriptions. Read-only and network controls vary by runtime.