Desktop app · macOS, Windows, Linux
Personal bots keep up with your day: briefs, research, reminders. Coding agents build in your projects. A bot can hand a coding task to an agent and bring the result back to you.
Powered by Claude Code, OpenAI Codex and Grok. Nothing to train or host.
Claude · Your project companion
Thread started in booking-api · Claude
booking-api · Claude
Review validation
Add boundary tests
− if (input.startAt === null) return invalid();+ if (!input.startAt?.trim()) {+ return invalid("Start date is required");+ }Each task gets its own thread, plan and progress.
Plans, diffs and test results sit next to the chat.
Approve a step, queue a follow-up or change direction.
How it works
Canopy is a desktop app for getting work done with AI. You ask, Claude, GPT or Grok does the work, and you decide what happens next.
1
Talk to a bot like you would to a colleague.
2
The bot reads the tools you connected.
3
A short brief, then you choose what happens next.
Coding works the same way. Open a thread, and an agent plans, edits and tests in your project.
Pair with AI every day without a wall of terminal windows.
Ship your own product with agents that write, test and review.
Get briefs, research and reminders from a bot that knows your work.
Bots and threads
A bot handles the ongoing things. A thread is where an agent works on code. A bot can start threads for you.
01 Bots
A bot remembers the conversation, checks your connected tools and follows your routines.
You
“Tell me when customers ask about refunds.”
Support bot
Two refund questions in #support today. I drafted replies.
On your schedule
Remembers the context Brings back the next step
02 Threads
An agent works on real code in your project. Add a side thread for a second agent’s review.
src/checkout.ts +2 −1
validate(address) {
− return !!address;
+ return address.trim()
+ .length > 0;
}Claude Fix implemented
Codex Reviewing edge cases
Storefront Workstation
Runs on your computer Follows your permissions
Your bot starts the thread, follows the work, and brings the decision back to you.
A day with Canopy
Stop copying context between terminals, tabs and chats. The conversation, the code and every decision stay together.
9:00 AM
Your bot checks GitHub and Linear on a schedule and tells you what needs you.
9:20 AM
Each task gets its own thread. They run side by side, each with a plan you can follow.
11:05 AM
Approve a step when the agent asks. Queue a follow-up without waiting for the turn to end.
4:30 PM
Your bot reminds you of what you asked it to track. Tomorrow, every thread is where you left it.
Example day with sample data.
Get started
Three steps. No Canopy account needed.
01
Download the desktop app and open it. Canopy sets up what it needs on first start.
02
Use Claude through Claude Code, GPT through OpenAI Codex, or Grok. Sign in or add API access. One is enough.
03
Describe the task. Add a project when an agent should work on code.
Canopy uses the official agent tools: Claude Code from Anthropic, Codex from OpenAI for GPT, and Grok from xAI. There is no model to train or host. Model access comes from your provider, and their plans and usage limits apply.
Examples
A morning check-in. A bug that needs a second look. A project left running at home. Pick an example and watch it play out, step by step.
01
Ask for the blockers, changes, and decisions that matter today. Your bot can check connected accounts and existing threads before investigating a question in a new thread.
Sample workflow Step 4 of 4
All steps shown · Reduced motion
Project bot
Weekdays at 9:00 AM · Storefront
Your request
“Every weekday at 9, brief me on Storefront. Check Linear and GitHub, then investigate any checkout blocker without changing files.”
Your bot
There’s a checkout issue and a failing check. I’ll ask a read-only thread whether they’re connected.
Workstation
StorefrontCoding thread
Claude traces the failed check. A Codex side thread gives the findings another look.
Claude
CodexSide thread
Investigation · Read only SHOP-128 → checkout validationFailing check → guest address case Finding: the same missing validationThe result comes back
The bot brings the investigation back into your morning brief.
No files changed. You decide whether to start the fix.
Illustrative examples, not live activity. Connected accounts need your grants. Threads follow the permissions you set.
Connected work
Morning triage
A new issue
Linear SHOP-128
Bug Checkout
“An empty address gets through validation.”
Your bot picks it up
Reads the issue.
Starts a coding thread.
Allowed in Storefront
Agents do the work
Workstation
You make the call
Needs your approval
The fix and review are ready.
Works with your connected accounts
Powered by remote MCP. Bring your own server for more tools. Microsoft 365 uses your organization’s endpoint.
Linked computers
Leave your agents working at your desk. Pick up the thread from wherever you are.
Claude read src/
Claude edited validate.ts+2 −1
− return !!address;+ return address.trim().length > 0;bun test checkout24 passed
Check guest checkout too.
From LaptopDelivered to Workstation
Codex Side thread
Reviewing guest checkout
Reply to this thread
WorkstationLinked computers
LinkedWorkstation
Home · 2 threads running
Laptop
This computer · Following
Shared between them
Stays on each computer
On your phone, open the web app when web access is configured.
Example thread with sample activity. Remote actions need the home computer to be reachable.
Mobile apps · Coming soonGo a little deeper
Pick the part of your day you want to change. Explore how the pieces work, what to connect, and where to start.
Questions
What Canopy is, what it needs and where it runs.
Read the setup guideNo. Canopy is a desktop app. It runs Claude Code, Codex and Grok, the official agent tools from Anthropic, OpenAI and xAI, and gives their work one place.
No. You use models like Claude, GPT and Grok through their official agent tools. Sign in to a tool, or add API access. Provider plans and usage limits apply.
Install the agent tool on your computer. Canopy can install it for you on supported platforms. Then sign in through the tool, or add API access in Settings → Providers.
Bots keep an ongoing conversation, use connected tools, and follow routines. Threads let coding agents work with files and commands in a project on your computer.
Yes. A bot can start a thread, send follow-up instructions, and report back when the work finishes. You choose what it can do. Approvals also appear in the bot conversation.
Linked desktop apps can share settings, providers, connectors, skills and plugins. They keep readable copies of bot and thread history. Project files and native CLI logins stay on their own computers.
Yes. Set a recurring routine or a reminder, and tell your bot when to message you. Scheduled work needs its home computer or server and assigned worker to stay available.
No. The desktop app runs Canopy on your computer and sets up its local services. The first start needs internet. You can also run Canopy on your own server. Either way, your provider processes model requests.
Get Canopy
Install the app. Connect a provider.
Give your first bot or thread something to do.