CALVRASA desktop coding agent that works with files, commands, and project context. Built to keep the process visible from the first instruction to the final check.
The desktop agent can read files, list directories, create new files, write content, and edit selected line ranges within a project.
The desktop runner executes shell commands in the project directory and returns their output, exit code, and duration to the agent.
The agent loop receives tool results, decides what to do next, and continues until it completes, stops, reaches its step limit, or encounters a blocker.
File actions produce change information, and command actions expose their output. The activity view separates successful and failed steps so you can review the result.
Select a project, follow recent work, and describe a task with its relevant files and constraints. Context and model quality determine how well the agent can respond.
Calvras integrates AI requests with services including Gemini, OpenRouter, and Hugging Face. Availability depends on configured credentials, providers, and usage limits.
Name the result, relevant files, and constraints. A focused task gives the agent a useful starting point.
Inspect the plan, tool activity, changes, and command output. A failed step should inform the next action.
Run the checks your project needs and review the result. Generated code still needs your judgment.
Desktop access. Native filesystem and command execution require the desktop runtime or an appropriate local connection. Visiting the public website does not grant access to your computer.
Network access. AI requests use external providers and may include prompts and selected project context. Local files do not imply that all AI processing happens locally.
Feature availability. Models, limits, and capabilities vary by configuration and runtime. The pricing page describes plan tiers; paid checkout is not yet connected.