Task and context

You describe a goal in the Bytos panel. Bytos reads the workspace, searches the repository and gathers the files, symbols and diagnostics it needs. The agent does not start by editing blindly; it builds context first.

Plan and approval

Bytos proposes a plan. The plan can be reviewed before execution. Reads, edits, terminal commands, browser access and MCP tools are approval-gated by default. Auto-approval is opt-in per category, not the default.

Approval is the core safety boundary. It is also why "autonomous" must be read as supervised autonomy: the agent can run a multi-step loop, but the developer controls which actions are allowed.

Execution and tools

Bytos edits files as reviewable diffs, runs commands in a VS Code terminal, can browse the web and can call MCP tools you configure. Checkpoints and task history provide a recovery path. You can stop a task and inspect what changed.

Verification

Verification is part of the loop: Bytos can run tests or commands and read the output, then revise its approach. Verification depends on the project having a runnable test or build command; the agent does not guarantee correctness. Review the final diff.

Sequential Auto Phase

Auto Phase lets you define up to ten phases. Each phase can have its own provider, model, prompt, attachments and MCP allowlist. Phases run in order, and a phase can pass a short completion report to the next. It is sequential orchestration, not a parallel multi-agent system.

Current platform

Bytos currently runs as a VS Code extension. JetBrains, Cursor, Windsurf, CLI and a standalone IDE are future targets, not current capabilities.

Install Bytos for VS Code

Bytos is currently a VS Code extension. Install it from the Marketplace, connect your own provider and start with an approval-first workflow.