Download Cody IDE
Pick your operating system. We auto-detect what you're using and recommend the right build.
Windows
Download for Windows (64-bit) Windows ARM64 (115 MB) .msi installer (enterprise / GPO deployment)Verify checksum (SHA256)
1fcbddfb04369699e7801df04fb7b50f876789dd91efd5dd3097902a6c179d54
macOS
Download for Apple Silicon (M1/M2/M3/M4) Intel Mac (238 MB)Verify checksum (SHA256)
0f6b6eb96d30db3724bc391b7e2abb6111bed56aa3f2e8bc6fea04d7779222f6
Linux
Download AppImage (recommended) .deb (Debian/Ubuntu) (108 MB) .rpm (Fedora/RHEL) (154 MB) .tar.gz (generic)These installers aren't digitally signed yet, so your OS may show a warning the first time you open them. It's safe to continue.
An IDE with an AI agent that works on your project
Cody IDE is built on the open source VS Code editor and ships with Cody, an agent that understands your code, plans, edits files and runs commands. The agent lives inside the editor, connected to your Cody account, and every important action goes through your approval.
Ways of working
Choose how the agent works depending on the task.
Ask
Ask anything about your code. It investigates the project with its tools and answers without changing anything.
Plan
Builds a plan or an architecture for what you need. The plan is saved only once you approve it.
Code
Makes real changes: creates and edits files, runs commands and checks the result.
Spec
Spec-driven development: requirements, design, tasks and code, stage by stage, with your sign-off on each one.
Team
A team of specialized roles coordinated by Cody (product, architecture, QA, security, frontend, backend and more) works the task in phases.
What the agent can do
Explore and search code
Reads files, searches the contents of the whole project and finds files by name, without you telling it where to look.
Create and edit files
Applies changes directly to your files, as targeted edits or whole files.
Run commands
Runs commands in the integrated terminal (tests, builds, installs) and uses the output to keep working.
Visible task plan
On multi-step tasks it keeps an up-to-date plan so you know what's done and what's left.
Inline completions
Code suggestions as you type, right in the editor.
Voice dictation
Talk to the agent instead of typing.
Session history
Pick up previous conversations with all their context.
Choose the model
Switch AI models depending on the task, among those available to your account.
MCP and integrations
The agent connects to your tools through MCP (Model Context Protocol). Turn on the connectors you need and the agent can query and work with those systems.
GitHub
Repositories, issues and pull requests, with your personal token.
GitLab
GitLab.com or your own self-hosted instance.
Jira, Confluence and Bitbucket
Atlassian tickets, documentation and repositories with your account.
Figma
Your designs as context, through the Figma Dev Mode MCP server.
Your own MCP servers
Add any MCP server (local or remote) from the project configuration.
Connected to the Cody platform
Your Cody account
Sign in with the same account you use on the web platform.
Acceptance Explorer
Open the analysis of the project you have open in the editor.
Sync your project
Send your project to the platform for analysis.
Agent Cells
Connect the IDE to your agent teams on the platform (Pro plan).
Usage at a glance
Your usage and available credits are shown above the chat.
Secure by design
An agent that edits files and runs commands needs clear limits. These are Cody IDE's.
Confined to your project
It can't read or write outside the folder of the open project.
Protected folders
It doesn't modify git hooks, CI workflows or editor tasks, which could run code on their own.
Your secrets stay unread
.env files, private keys and credentials are out of its reach.
Commands need your approval
Every command is shown before it runs, and data exfiltration patterns are blocked.
External content is data
Whatever comes from files, issues or external pages is treated as information, never as instructions for the agent.
Per-tool permission
Each tool from an MCP server asks for your authorization before its first use.
Pauses on long tasks
On very long jobs the agent stops periodically and asks whether to continue.
Review before applying
With autopilot off, changes stay pending until you accept or reject them.