Completely replace the system prompt of any built-in agent, including the lead. You decide how it should investigate, write code, and report the result.
Save your prompt in a Markdown file. Use a global definition or one for a specific project, without changing Clarvis itself.
01You maintain this project.
0203Read the contribution guide before editing.
04Follow the existing code style.
05Run tests for the code you changed.
0607When you finish, explain what changed
08and how you checked the result.
The file body replaces the whole built-in prompt. Tool permissions are configured separately.
Marshall, Admiral, Coder, Explorer, or Planner. Your choice.
02 / Extension Profiles
A profile for each kind of work.
One profile with only research extensions. Another for coding. Another for review or whatever you need. Each loads only the plugins and skills you selected.
Fewer input tokens. Tools from plugins outside the profile, including their MCP servers, are not presented to the model. Fewer tool definitions in context can lower usage costs.
1
Choose
Give a set of installed plugins and skills a name.
2
Review
See the agents, tools, and automation included before activating.
3
Switch
Switch profiles as the task changes, even within the same project. Use global profiles across projects.
See the plan, file changes, tool results, and tests in one terminal. Send a follow-up while the agent works, or stop it when you need to.
ClarvisProduct recording
A real Clarvis run using Qwen3.8-Max to plan, implement, inspect failures, and verify the result. The recording is edited for pace; the product output is unchanged. Read a summary of the video
Clarvis receives a task and creates a plan.
The agent reads the project and delegates parts of the work.
You can review file changes as the agent works.
A check fails. The agent investigates and fixes the problem.
Clarvis runs the checks again and explains the result.
Providers, gateways, and local endpoints
OpenAI
Anthropic
Qwen
DeepSeek
MiniMax
Google
Mistral
xAI
OpenRouter
Ollama
Hugging Face
NVIDIA
03 / Models
Choose the models you want to work with.
Connect an API, a supported subscription, or a local server. Choose a lead model and, if you want, different models for your sub-agents.
Subscriptions
Connect an eligible ChatGPT or Grok subscription. Access depends on your provider and account; this is not a provider endorsement.
Provider APIs
Configure native OpenAI, Anthropic, or Google connections.
Other services
Bring a private gateway or an OpenAI-compatible service.
Local models
Point Clarvis at a local compatible server and define its model limits.
