BoltchatBoltchat combines a Monaco code editor with an AI agent that can plan work, inspect a project, propose file changes and run relevant checks. You stay in control of the folder, commands and edits.
A local workspace around your project
Real editor
Tabs, project search, syntax highlighting for more than 80 languages and the editing engine used by Visual Studio Code.
Task planning
The agent turns a request into visible steps and updates its plan as work progresses.
Reviewable edits
AI changes appear as diffs that you can inspect before accepting, with undo for accepted work.
Verification
The agent can run tests, inspect console output and use a built-in browser to check web projects.
How a coding task works
- Open and trust a folder. Boltchat only gives the agent project access after you explicitly select and trust that folder.
- Describe a concrete outcome. Ask for a bug fix, feature, refactor, explanation or test. The agent creates a plan before making broad changes.
- The agent reads what it needs. It searches and opens relevant text files rather than sending the entire project automatically.
- Review proposed changes. File edits are shown as additions and removals. You can accept, reject or undo them.
- Check the result. With permission, the agent runs the project's tests or build and can inspect a local web page.
Review, branches and undo
AI edits do not silently replace the conversation that produced them. Edited messages and regenerated answers retain earlier alternatives as branches, and file checkpoints support undo after accepted changes. You should still keep normal project backups or version control for important work.
Your responsibility: AI-generated code can be wrong, insecure or incomplete. Read meaningful diffs and run the project's own checks before shipping.
Command safety
The coding agent cannot run arbitrary commands without a gate. Boltchat checks every proposed command, blocks clearly dangerous patterns and can ask for your approval. Settings let you require approval every time or use the smart gate for routine project commands.
- Commands run with your Windows account inside the opened project folder.
- Environment variables whose names look like keys, tokens or passwords are removed before an agent command runs.
- Command output is shown in Boltchat and sent to Groq so the agent can interpret the result.
- You should decline any command you do not understand.
A real integrated terminal
The workspace includes PowerShell or Command Prompt tabs rooted in the project folder. Your terminals and the agent's command output remain visible beneath the editor, so you can reproduce checks or continue manually.
Browser-based QA for web projects
When enabled, the agent can open local project pages or a development server in Boltchat's built-in browser, interact with the page, read console messages and ask a vision-capable model to describe a screenshot. Opening unrelated external websites requires your approval.
What leaves your PC
Your project remains on your PC, but content must be sent to Groq when the model needs it. That can include the text of files the agent reads, your selected code, filenames, search queries, proposed commands, command output, browser text and screenshots requested for visual checking. Boltchat does not operate an intermediary server.
See Code workspace privacy and terminal and browser privacy for the precise policy.
What the agent cannot promise
- It cannot guarantee that a change is correct merely because tests pass.
- It cannot use tools that are not installed on your computer.
- It should not be given access to folders containing secrets or unrelated sensitive data.
- It does not replace code review, backups, version control or security testing.
Ready to try it? Follow the Boltchat installation guide.