CodeVista

Agent workflow

Plans the work, calls the right tools, then reviews the diff.

CodeVista works like a local coding agent, not a single-shot chatbot. It breaks a request into steps, reads and edits files, runs shell commands, searches the codebase, and checks what changed before handing work back.

Plan, act, observe, review
The tools you'd use yourself

Animation of a looping agent lifecycle: plan, act, observe, review.Plan, act, observe, review

Animation of the agent invoking file read, shell, code search, and web fetch tools.Tool calls

Codebase intelligence

Reads beyond the open tab and remembers project decisions.

The agent builds useful context from the repository instead of flooding itself with every file. It can identify relevant code, preserve project notes, and carry decisions across sessions so future tasks start with better context.

Reads the right files
Memory that outlives a session

Animation of a file tree being scanned and matched by a semantic retriever.Codebase retrieval

Local control

Works in your browser, your terminal, and your IDE.

CodeVista is the same agent on every surface. Open a sandbox in the browser, run it in your terminal, or install the extension in your IDE — the agent, tools, and permission model are identical. File writes, shell commands, and external calls always ask before they act.

Web, CLI, or IDE — your choice
Permission before action

Animation of three clients: VS Code, terminal, and browser cycling into focus.Developer surfaces

Animation of permission prompts cycling between allow, confirm, and deny states.Scoped permissions

What it does.

Plans the work, calls the right tools, then reviews the diff.

Reads beyond the open tab and remembers project decisions.

Works in your browser, your terminal, and your IDE.

Ready to try CodeVista?