Terminals that remember what they were doing. Claude Code, Codex, OpenCode — or CepelCode's own CLI — in real terminals you can name and give roles, so they work as a team. Edits that arrive in the editor typed, word by word, the way the agent wrote them.
Free · x64 installer · SHA-256 published with every release
Three minutes in
Home asks what to build, and hands it to a team or to any CLI you have — before a single file is open.
Run Claude Code, Codex, OpenCode, Gemini, Aider — whatever you already have — or CepelCode's own CLI, on a free gateway or your own API key. The editor recognises each one.
Every file the agent touches opens and is typed in front of you, highlighted when it lands. Stop it, checkpoint it, undo it.
New in 1.8.0
Give a terminal a name and a role — Ava the Leader, Ben the Developer, Cleo the Tester — and the agent in it knows who it is and who else is on the team. Hand the leader a goal: it briefs the team, asks what each is good at, splits the work into tasks, assigns them, checks what comes back, and works too.
Built in · updated in 1.8.1
Type cepelcode in any terminal of the app, or press CepelCode under Launch. Run it free on FreeLLMAPI or OmniRoute — gateways on your own machine that pool the free tiers of dozens of providers behind one key — or on your own key for OpenAI, OpenRouter, Google Gemini, DeepSeek, xAI, Groq or Mistral. /connect sets it up; /model picks the model.
Dispatch
Pick one of 66 specialists, say what it should do, choose Claude Code, OpenCode, Codex or Gemini — whichever is installed — and a new or existing session. CepelCode opens a terminal there and starts your CLI as that specialist, with the task already given.
PS> claude --plugin-dir ~\cli-kit\claude\cepelcode --agent frontend-engineer "Add a pause menu to the HUD"
# Claude Code starts as @cepelcode:frontend-engineer
# every skill pack loaded as a skill; the task already given
# the pane is named for it and counted as an agent
The skills library
Thirteen skill packs ship with the app — templates, palettes, surface styles, type, motion, launch and mobile checklists — and every CLI running in CepelCode can read them. Ask for a landing page and the agent starts from one that already works.
OpenCode was asked for a landing page for a coffee subscription, twice. Counted in the two files it wrote:
Both pages looked finished. The difference is what you could publish.
Three modes, one switch
A dashboard, an editor and a wall of terminals want different layouts. CepelCode gives each its own mode and puts the switch dead centre — the same switch you see below.
One box for the task. Hand it to a team or to any CLI you have, and see what is already running — before you open a single file.
Files, tabs, breadcrumbs, diagnostics while you type, one terminal under the code. The editor you already know how to use.
Sessions down the side, terminals tiled across the whole window. For when the terminals are the work.
Built for agents
A file an agent changes is not snapped into place. It arrives the way it would have been written — cut lines go at once, new text word by word under a moving cursor, the change highlighted when it lands. Never dirty while it types; always the file on disk when it stops.
Claude Code, Codex, Gemini, Aider — run them the way you already do. The editor recognises each one, resumes it after a restart with its own --continue, and counts it as an agent. OpenCode gets a one-click launcher on top: it installs Node.js if missing, installs or updates OpenCode, and runs it, every step on screen.
A restored terminal is a record, not a process: name, folder, what it last printed — and no shell until you press a key. Twenty terminals cost twenty emulators and zero PowerShells.
The chat dock is a real PTY with a composer instead of a prompt line and a transcript instead of a scrolling buffer. It keeps its own width, and it has tabs.
Compiled into the Rust core. No model, no key, no network. They read one sweep of your workspace and their findings land as Problems rows and squiggles — on every save.
Exit codes, the working directory and what actually ran come from the shell's own markers, not from reading the screen. Listening ports come from the kernel's table.
CtrlAltSpace from anywhere. Its own recogniser, compiled in — no download, no network — biased toward your codebase's identifiers. Clear words with small pauses; it says what it cannot do rather than pretending.
Checkpoint the workspace before an agent starts, review what it changed, undo the last run, keep agents in their own lanes, and check the tree against rules you wrote down.
The update chip shows its state at a glance — available, a download filling the chip, ready to restart. Clicking it shows the release notes for every version you are about to get, and after an update a "What's new" sheet appears once.
Invite people with a code; the host lets them in. Activity keeps a live card for each of them in their own colour — the file they have open, their line, the code around their cursor as they type, whether an agent is writing.
The harness
Driven over whichever coding CLI you already have. A router reads your request and decides who is needed — and its most common, most correct answer is nobody, do it inline. It does not spawn a fleet by reflex.
Built to be trusted
Checks for a new version at launch and every six hours. On your say-so it downloads a package of what changed, verifies it, swaps it in after the app closes and starts itself again — no installer, nothing new for Smart App Control to block. Thirty days after a release, the old version asks you to update or quit, and lets you stop your running agents first.
The only network call the app itself makes is that update check. Analysis runs in a local Rust core. Dictation never leaves the machine. Your agents talk to whichever provider you configured in your CLI.
Every run carries a capability token the app minted for that session. A token copied to another machine fails even though the bytes are right. A person at the keyboard is trusted; a cron job nobody meant to point here is not.
Download
Windows 10 or 11, x64. One installer, no account.
SHA-256
What's new
Questions
Yes — download it and use it. There is no account, no trial clock and nothing to unlock.
For an unsigned installer, yes: click More info → Run anyway. Every release publishes its SHA-256 beside the file so you can check the bytes you got are the bytes we built. Signing is on the roadmap.
No. The app's only network call is the update check. Analysis runs locally in a Rust core; dictation runs locally. Your agents talk to whichever provider you configured — and CepelCode's own CLI sends what you ask, and the files it reads, to the gateway or provider you connected and nowhere else.
Any that runs in a terminal. Claude Code, Codex, Gemini and Aider are recognised in the terminals, resumed after a restart with their own --continue, and driven by the harness. Claude Code, Codex and OpenCode are also connected to the app itself, so they can read its diagnostics and join a team. OpenCode has a one-click launcher, and CepelCode's own CLI is built in.
A coding agent built into the app: type cepelcode in any of its terminals. The CLI itself is free and needs no account. It runs on a model you connect: FreeLLMAPI or OmniRoute — free gateways on your own machine that pool the free tiers of dozens of providers behind one key — or your own API key for OpenAI, OpenRouter, Google Gemini, DeepSeek, xAI, Groq or Mistral. Keys are stored on your machine.
FreeLLMAPI: install its desktop app from its GitHub releases, add free provider keys on its Keys page, then copy the freellmapi-… key from the page header (or Copy Key in its tray icon). OmniRoute: run npm install -g omniroute and then omniroute, add providers at http://localhost:20128 under Providers, then copy your key from Dashboard → Endpoints. Either way, run cepelcode connect and paste it — the CLI opens the dashboard for you.
Give terminals a name and a role from their title bar, or start a team from Home. Each agent is told who it is and who else is on the team; they message each other and keep a shared task board through CepelCode. The leader assigns and reviews, and you can message any of them from the Team panel. It works with Claude Code, Codex, OpenCode and CepelCode's own CLI.
No. They are independent products of their own makers, with their own licences, accounts and pricing. CepelCode is an editor that runs them in its terminals; it is not affiliated with, endorsed by or a reseller of any of them, and it never sees the traffic between a CLI and its provider.
Yes, inside CepelCode. OpenCode started in any of its terminals lists every skill pack and every specialist; Dispatch starts Claude Code with them loaded as a plugin; and any CLI connected to CepelCode can list and read them through its tools. They live in CepelCode's own profile. The one thing it adds to Claude Code's and Codex's own settings is its connection, with its read-only and team tools allowed; outside the app that connection finds nothing and offers no tools.
Not for the editor, the terminals or the analysers. The OpenCode button installs Node.js for you if it is missing. CepelCode's own CLI uses Node.js for its full interface and falls back to a plain line mode without it. Only the agent harness needs Python 3, and it tells you in plain words if it is not there.
Windows first. The core and the app are portable; other platforms follow once Windows is solid.
Feedback
A bug, a wish, a thing that annoyed you. It goes straight to the person who builds CepelCode.