Orvlet preview

product

The workbench

One workspace for long sessions: talk to an agent, read its diffs, run the terminal, review the pull request — without handing your project to a hosted service.

CSS illustration. Chat and Code are two views of the same workspace; the editor keeps the full width in Code view.

What the agent can do — with your say-so

Code

Reads, edits and writes files; applies atomic multi-file patches; structural search-and-replace with ast-grep; language servers for diagnostics, definitions and renames. Reads .docx, .xlsx and .pptx; creates documents, PDFs and slides.

Commands

Runs shell commands in an integrated terminal under the macOS-enforced command sandbox. Interactive terminals, background tasks and loops that repeat until a check passes — always pausing for approvals.

Browser & computer

Browser Use in Orvlet's built-in browser with a separate profile per task. Computer Use (macOS-only, off by default) can operate other apps through a bundled helper with per-app approvals, a visible indicator and a global stop shortcut.

Knowledge

Knowledge collections index folders, files and PDFs on this machine — scanned pages by OCR — with cited search results for you and the agent. Secret files are never indexed. More on documents.

Git

Worktrees, a review pane with per-hunk stage and revert, GitHub pull requests with comments pulled back into the task, and multi-folder projects whose identity stays their main folder.

Teams of models

Multi-agent commands such as /boost, /research and /review, dynamic workflows, subagents, checkpoints, scheduled tasks and model Rooms where models talk to each other. Helper agents follow the parent task's permission mode.

Stars over a black desert ridge

Approvals show the exact command, URL or diff — before anything runs.

Photograph — David Billings via Unsplash

Desktop, web and remote

The desktop app targets macOS 27+ on Apple Silicon. An optional self-hosted web UI and SSH/Docker remote workspaces exist for projects on other machines; neither is needed to start the desktop app. Remote control from a phone or browser is served by the desktop app itself — off by default, no relay.

Command-line interface

The orvlet CLI runs tasks headlessly, reviews diffs, manages MCP servers and follows the engine log. Without --mode it runs in Ask before changes, and anything that would ask is denied because nobody is there to answer.

orvlet --prompt "Fix the flaky test"
orvlet review
orvlet engine logs --follow --stats

A full editor, terminal and Git client

The workspace is built for reading and changing code yourself, not only for watching an agent do it.

  • Language intelligenceCompletion, hover, go to definition and references, rename, code actions and diagnostics through language servers.
  • Real debuggingBreakpoints, launch and attach, step, call stack, variables, watches and debug output through the Debug Adapter Protocol.
  • Test ExplorerDiscovered tests, run or debug one or all, structured failures you can jump to, and coverage.
  • Editing comfortsMulti-cursor and column selection, Vim mode, code folding, minimap, soft wrap, inline color previews and find and replace.
  • Inline AIGhost-text completion, next-edit suggestions across dependent files, inline prompt edits and one-click fixes for diagnostics.
  • TerminalTabs and split grids, semantic command blocks, search, bookmarks, persistent shells, and a fix suggestion when a command fails.
  • Visual GitCommit graph, blame gutters, stash browser, reflog recovery, interactive rebase, cherry-pick dry runs and a three-way merge tool.
  • Branches and worktreesWorktree per task, dependent branch stacks, submodules, Git LFS and an undo journal for repository operations.
  • ReviewPer-turn diffs, line comments, per-hunk stage and revert, and a selection you can hand straight to the agent.
  • PreviewDev servers with hot reload, multiple preview tabs, color scheme emulation and an Android emulator pane.
  • Layout your waySplit main and side areas, pop panes out into their own windows, and restore them after relaunch.
  • Code mapsA Tree-sitter repository map ranked by importance, symbol graphs and compact task context packets.

Explore every area

Orvlet does a lot. Each area has its own page.

Agents & automationA personal background agent, subagents, workflows, schedules and loops. Browser & computer useA built-in browser for the agent and approved control of Mac apps. Documents & researchOffice files, PDFs, knowledge collections, cited research and data. Voice & studioDictation and conversation, images, music, audio, video and design. Remote & integrationsSSH and containers, your phone, live sharing, editors and the CLI. Orvlet EngineNative MLX and GGUF inference with caching and speculative decoding.