Koda Docs
Home MCP Download
Docs / Features

Features

Everything Koda gives the agent CLIs — organized, persistent, and designed for running many of them at once. This page is a reference for each surface in the app and how to use it.

Workspaces

A workspace is any folder you open in Koda. It becomes a self-contained environment: your file tree, editor tabs, terminal layout, open panel, browser URL, and task board all belong to that workspace and survive restarts.

Organizing workspaces

The left switcher groups your projects into three categories you can drag between:

Each workspace can be given a custom display name (inline rename) and reordered by dragging. Selecting a workspace does not reshuffle the list.

Persistence & templates

Layout and session state are written to <workspace>/.koda/session.json (debounced) and restored when you reopen. Open files survive in-session workspace switches; on a full restart, files are re-read fresh from disk and terminals respawn.

New workspaces can start from a template — Empty, Web app, Python/ML, or Claude feature dev — which seeds a few Kanban cards, opens the right panel, and can suggest a theme. Templates only write to .koda/; they never touch your project files.

The .koda/ directoryKoda keeps all of its per-workspace state (session, board, coordination, env) in a .koda/ folder that is automatically gitignored. Your repository stays clean.

Terminal panes

Terminals are the main stage. Each pane is a real shell (a PTY via node-pty), not a fake console — anything you'd run in your normal terminal runs here.

Agents

In Koda an agent is a launch command run inside a pane. Koda ships built-in profiles and lets you define your own.

Built-in agents

AgentProviderNotes
Claude CodeAnthropicSign in with the real Claude Code CLI, or inject an API key.
OpenAI CodexOpenAISubscription via CLI login, or API key.
GeminiGoogleVertex AI / AI Studio API key.
GrokxAISubscription via CLI login, or API key.
opencodeanyAny CLI-based agent.

Custom agent profiles

Open Manage agents… (from the new-pane menu or the command palette) to define your own agent: a name, a launch command, an optional binary to detect, a provider (for vault key injection), an accent color, and a glyph. Custom agents are stored globally so they're available in every workspace, and they appear in both the new-pane menu and the Kanban agent picker.

Koda auto-detects which agent binaries are installed and only offers the ones it finds.

Kanban task board

The Tasks tab is a board where cards are work, not sticky notes. It's the engine that launches agents.

The board is also the coordination backboneWhen coordination is on, every agent session also creates and updates its own card here automatically — see Coordination (MCP).

Monaco editor

The same editor core as VS Code, docked beside your agents in the Editor tab. Use it to review and tweak what your agents write without leaving Koda.

Single-file language servicesThe editor provides per-file highlighting and editing, not a cross-file language server — there's no project-wide IntelliSense or go-to-definition across files. For deep code intelligence, let the agents do the heavy lifting.

File tree

A lazy, recursive file tree lives as a slim strip inside the Editor tab, scoped to your workspace. Click a file to open it in a new editor tab and focus the editor. The tree hides dotfiles, so .koda/ and other dot-directories stay out of view.

In-app browser

The Browser tab is a real embedded browser pointed at your dev server — preview your app right next to the agents building it, with no alt-tabbing.

Links that open new windows are sent to your OS browser, and the embedded view is locked down (no Node access in the guest content) for safety.

Credential vault

Store API keys once, encrypted by your operating system. Tag a key with a provider, and Koda injects it into that provider's panes at launch.

Encryption is machine-boundBecause the vault uses OS-level encryption tied to your user account, secrets won't decrypt under a different OS user or on a different machine. Re-enter keys if you move to a new device.

Themes

Koda ships 30+ built-in themes, dark and light, all hand-tuned. One click re-colors the entire app live — the UI, your terminals, and the Monaco editor all at once. Favorites include Tokyo Night (the default), Catppuccin, Dracula, Monokai, Solarized, One Light, and GitHub Dark.

Open the theme gallery from Settings → Appearance (the header gear) or the command palette; your choice persists across restarts.

Per-pane accounts

Run different logins in different panes. Assign a specific Claude account per terminal, so one pane uses your work subscription and another your personal — with no re-auth dance. Manage logins via Manage accounts… in the new-pane menu.

Under the hood each pane gets its own config directory, so accounts stay fully isolated between panes.

Environment variables

Inject custom environment variables into agent panes at spawn, from two sources that share one path:

When both define the same variable, the agent's value wins. Variables are applied at spawn, so reopen a pane to pick up edits.

Reserved variablesKoda manages KODA_PANE_ID, KODA_COORD_DIR, and CLAUDE_CONFIG_DIR itself — these can't be overridden by your env entries.

Settings

Everything that used to live in scattered header buttons now folds into one place: a full-page Settings surface with a left tab rail. Open it from the gear in the header or via the command palette.

TabWhat's there
ConnectionsConnect Claude Code, manage Claude accounts, and see which agent CLIs Koda detected on this machine.
MCPWire up external tool servers — see MCP Connections.
AppearanceThe theme gallery — all 30+ palettes, applied live.
CredentialsThe encrypted credential vault for API keys.
AccountYour Koda account and plan tier — sign in to unlock Pro.
AboutVersion, update status with a one-click Restart to update, and links.

Account sign-in also lives at the bottom-left of the app, always in reach, showing your email and tier badge once you're in. No account is required to use Koda — signing in only unlocks paid features like code review.

Command palette

Press Ctrl/Cmd+K anywhere — even over a focused terminal — to open the command palette. Fuzzy-filter across grouped commands: open a new pane for any agent, create or switch workspaces, change the theme, open the vault, manage agents or accounts, and more. There's also a header button if you prefer the mouse.

Seamless updates

Koda keeps itself current. It checks GitHub Releases on launch and every 30 minutes, downloads updates in the background, and shows a status pill in the header. When an update is ready you get a Restart to update toast — one click installs it and relaunches. Update activity is logged locally so issues are diagnosable in packaged builds.

Usage meter (optional)

An optional header pill that shows your official Claude.ai usage — the real 5-hour and 7-day limit utilization and reset times, read straight from Claude.ai. It's there so you can see how much headroom you have before you start a long agent run, without leaving Koda.

It is strictly opt-in. Nothing happens until you paste your Claude.ai sessionKey once; with no session stored, the pill is absent and Koda makes no usage requests at all. You can disconnect it at any time, which deletes the stored session.

The one place Koda reads a provider credential — by design, and only if you askEverywhere else, Koda runs the real CLIs and never touches your tokens (see How Koda stays compliant). The usage meter is a deliberate, opt-in exception: there is no compliant local source for the official limit percentage, so — only when you choose to connect it — it reuses your own Claude.ai web session to read your own usage from your account on your machine. We surface that trade-off plainly rather than hide it. If you'd rather Koda never read a provider session, simply don't connect the meter.