Koda documentation
Koda is a desktop agentic development environment (ADE) — a workspace built around coding agents. You run the real agent CLIs in tiled terminals, launch their work from a Kanban board, and they coordinate with each other automatically. This guide covers everything from install to the multi-agent coordination layer.
Install Koda
Download the desktop app and open your first workspace.
Start →Explore features
Panes, agents, board, editor, browser, vault & themes.
Features →Run Autopilot
Hand Koda a goal and let it build, test, and loop until done.
Autopilot →Coordination / MCP
How agents auto-announce, claim work, and stay in sync.
Learn →KodaMemory
A shared knowledge graph so agents recall what the last one learned.
Learn →AI Second Brain
Koda learns who you are and briefs every agent — outputs tailored to you.
Learn →Introduction
Koda is a terminal/CLI orchestrator. Each pane in Koda is a real PTY shell, and an "agent" is simply a launch command — claude, codex, gemini, and others — run inside one of those shells. Koda gives those agent CLIs a polished, organized home: tiling workspaces, a built-in editor and browser, an encrypted credential vault, a task board that launches agents, and a coordination layer that lets many agents work together without colliding.
Koda is built on Electron, React and TypeScript. It runs entirely on your machine — your code, your shells, and your agents never touch a Koda server.
Installation
Koda is a desktop application with installers for Windows, macOS (Apple Silicon & Intel) and Linux. The Windows build updates itself in the background.
- Download the installer for your platform from the releases page — Windows
.exe, macOS.dmg(arm64for Apple Silicon,x64for Intel), or Linux.AppImage/.deb. - Run the installer. On Windows it's
Koda-x.y.z-setup.exe(assisted install — pick a location or accept the default). On macOS, open the.dmgand drag Koda to Applications; the build is unsigned for now, so the first launch needs a right-click → Open. On Linux, mark the.AppImageexecutable and run it, or install the.debwithapt/dpkg. - Launch Koda and sign in — creating an account starts your 7-day free trial with every feature unlocked, no credit card required. After the trial, a Pro subscription keeps Koda running.
You also need at least one agent CLI installed and on your PATH — for example Claude Code (claude), OpenAI Codex (codex), Gemini, Grok, or opencode. Koda detects which ones are present and offers them when you create a pane.
First run
The first time you open Koda, an onboarding flow walks you through four quick steps:
- Welcome — a short tour of what Koda is.
- Pick a theme — choose from 30+ palettes; the whole app recolors live.
- Providers — Koda detects installed agent CLIs, lets you sign in to Claude Code, or paste an API key straight into the encrypted vault.
- Create a workspace — open a folder and optionally start from a template (Empty, Web app, Python/ML, or Claude feature dev) that seeds a few Kanban cards.
After onboarding you land in your first workspace: the terminal grid is the main stage, with a collapsible side panel on the right holding the Browser, Editor, and Tasks tabs.
Core concepts
A handful of ideas show up everywhere in Koda. Understanding them makes the rest of the docs click.
Workspace
A folder you've opened. Holds your panes, layout, board, and session — all persisted under a local .koda/ directory.
Pane
One tile in the terminal grid: a real shell. Resizable, renameable, and the home of a single agent.
Agent
A launch command run inside a pane (e.g. claude). Built-in profiles plus your own custom ones.
Card
A unit of work on the Kanban board. Carries a title, a prompt, an assigned agent, and a status.
Lane
An agent's declared area of responsibility (e.g. backend · src/api) so peers stay out of each other's files.
Coordination
The MCP bus that lets agents read the board, claim work, and message each other. Included in the trial and Pro; auto-enabled for each workspace you open (except those you mark Sensitive).
Memory
KodaMemory — a shared knowledge graph. Agents record decisions, gotchas and pointers as linked nodes, and new sessions recall what's relevant to their task. Included in the trial and Pro.
Second Brain
Your personal context graph — Koda learns who you are (projects, goals, preferences, writing style) and briefs every agent, in every workspace, with the facts relevant to the task. Included in the trial and Pro.
Autopilot
Hand Koda a goal and it builds it — planning the board, working tests-first, and looping until your verify command passes.
Review
A diff of everything an agent changed since it started, with line comments and an Approve / Request-changes verdict. Included in the trial and Pro.
Connection
An external MCP tool server (docs, search, analytics…) wired into your workspaces so agents can call it.
How Koda stays compliant
Koda runs the real first-party CLIs in real terminals. This is a deliberate, locked architectural decision and it shapes everything:
- Koda does not make API calls on your behalf.
- Koda does not proxy, read, or reuse your OAuth tokens.
- Koda does not sit between you and your provider.
You sign in to your tools exactly as you always have — through the CLI itself. Because Koda only spawns the real CLI, using your own subscription inside it is ordinary individual usage. Your credentials never leave your device, and the coordination layer is a local tool server the CLIs choose to call — never a token proxy.
There is exactly one opt-in exception, and we name it openly: the optional usage meter. If — and only if — you choose to connect it, it reuses your own Claude.ai web session to read your own usage numbers. It's off until you connect it, the session stays OS-encrypted on your machine, and you can disconnect it anytime. Nothing else in Koda reads a provider credential.