Get started
Install Trove
Updated 12 Nov 2026
On this page
For agents
Load this page when: you are installing Trove for the first time or connecting a new agent
- In GitHub Copilot CLI, Trove installs a skill, four sub-agents, a session hook and an instructions block under ~/.copilot or $COPILOT_HOME.
- Coding agents use the CLI installer. Claude and Grok on the web use the MCP server instead, and ChatGPT installs Trove from its app directory.
- Check which agents are set up with
trove setup --status --json, and change them withtrove setup --enableor--disable. - On SSH, containers and other machines without a browser, trove login prints a link to approve on any device; in CI set TROVE_API_KEY.
- Remove Trove from a computer with trove uninstall; the account, books and API key stay.
Coding agents install through one command. Claude and Grok on the web connect through the MCP server instead, and ChatGPT adds Trove from its app directory, so none of them needs a local install.
Coding agents (CLI)
Run the installer in your terminal. It downloads trove, verifies its checksum, adds it to PATH, and then runs trove setup.
curl -fsSL https://heytrove.ai/install.sh | shOn Windows, run powershell -c "irm https://heytrove.ai/install.ps1 | iex" in PowerShell. The shell command above also works in Git Bash, which is where Claude Code runs its commands on Windows: it hands over to the PowerShell installer. On WSL, run the shell command above inside WSL. See Windows for what is different there.
The installer adds ~/.local/bin to your PATH in your shell's startup file: .zshrc for zsh, .bash_profile on macOS or .bashrc on Linux for bash, ~/.config/fish/conf.d/trove.fish for fish, and .profile for any other shell. Run it again from another shell and it moves that line to the new shell's file.
Setup signs you in and installs the Trove plugin into every agent it detects. On a desktop it opens the browser. Over SSH, in a container or without a display it prints a link and a code to approve on any device. In CI it skips sign-in and finishes signed out. In Claude Code, it also adds a session-start hook and allow rules for seven trove commands, so your agent can run them without a prompt each time: items list, items toc, items get, ms list, ms show, marketplace browse and marketplace subscribe. The last one lets the agent add a marketplace book to your library when a task needs it; remove a book again with trove marketplace unsubscribe. Every other command, such as changing a page of one of your books, asks you first.
You install and sign in one time. After that, you tell your agent what to do, and your agent runs Trove.
The installer sends a few install events: a random install id, the step, your operating system, CPU and the Trove version. When a step fails it also sends a short reason, and the end of the install command's error output with your home folder name removed. It prints a line saying so before it sends the first one, and the id is linked to your account when you sign in. To send none, run it with DO_NOT_TRACK=1. Then neither the installer nor trove sends any event, and no install id is saved.
curl -fsSL https://heytrove.ai/install.sh | DO_NOT_TRACK=1 shOn Windows: $env:DO_NOT_TRACK = '1'; irm https://heytrove.ai/install.ps1 | iex. TROVE_NO_TELEMETRY=1 works the same way.
If a download fails, the installer tries again up to two more times before it stops.
Supported platforms
| Platform | Supported |
|---|---|
| macOS | Apple silicon (arm64) and Intel (x64) |
| Linux | x64 and arm64 with glibc. Alpine and other musl systems are not supported. |
| Windows | x64. On Windows on Arm the installer uses the native arm64 build, or the x64 build under Windows 11 emulation when a release has no arm64 build. |
| WSL | Use the macOS and Linux command inside WSL. |
If your company blocks piping a script into a shell
You can download and check the files yourself, then run them. Every release lists its files and checksums in a manifest.
| Manifest key | Archive |
|---|---|
darwin-arm64 | trove-aarch64-apple-darwin.tar.gz |
darwin-x64 | trove-x86_64-apple-darwin.tar.gz |
linux-x64 | trove-x86_64-unknown-linux-gnu.tar.gz |
linux-arm64 | trove-aarch64-unknown-linux-gnu.tar.gz |
windows-x64 | trove-x86_64-pc-windows-msvc.zip |
windows-arm64 | trove-aarch64-pc-windows-msvc.zip |
On macOS and Linux:
V=$(curl -fsSL https://downloads.heytrove.ai/cli/latest.json | sed -n 's/.*"version"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')
curl -fsSL "https://downloads.heytrove.ai/cli/v$V/manifest.json"
A=trove-aarch64-apple-darwin.tar.gz
curl -fsSLO "https://downloads.heytrove.ai/cli/v$V/$A"
curl -fsSLO "https://downloads.heytrove.ai/cli/v$V/$A.sha256"
shasum -a 256 -c "$A.sha256"
tar -xzf "$A"
TROVE_INSTALL_ARCHIVE="$A" ./trove install "$V"Set A to the archive for your platform. On Linux without shasum, use sha256sum -c. trove install checks the archive against the manifest again, copies trove to ~/.local/bin, sets up your shell and runs trove setup. You can then delete the downloaded files.
On Windows, in PowerShell (on Windows on Arm, set $A to trove-aarch64-pc-windows-msvc.zip):
$V = (Invoke-RestMethod https://downloads.heytrove.ai/cli/latest.json).version
$A = 'trove-x86_64-pc-windows-msvc.zip'
Invoke-WebRequest "https://downloads.heytrove.ai/cli/v$V/$A" -OutFile $A
Invoke-WebRequest "https://downloads.heytrove.ai/cli/v$V/$A.sha256" -OutFile "$A.sha256"
(Get-FileHash $A -Algorithm SHA256).Hash.ToLower() -eq (Get-Content "$A.sha256").Split(' ')[0]
Expand-Archive $A -DestinationPath "trove-$V"
$env:TROVE_INSTALL_ARCHIVE = (Resolve-Path $A).Path; & ".\trove-$V\trove.exe" install $VThe checksum line must print True. To skip trove install, put the trove binary on your PATH yourself and run trove setup.
Windows
Trove runs on Windows x64 and Arm. A few things work differently there:
| What you see | What to do |
|---|---|
Windows blocks trove.exe | The message is "An Application Control policy has blocked this file", or a SmartScreen warning. The program is not code-signed yet. In a SmartScreen warning, choose More info, then Run anyway. Smart App Control has no per-program exception, so on a machine where it is on, use Trove through WSL, or through Claude or ChatGPT on the web. |
trove is not found after the install | The installer adds trove to your user PATH. Open a new terminal, and restart your coding agent so it reads the new PATH. |
| No "Trove is active" in Claude Code | The session hook needs the jq program. Ask your agent: "Install jq with winget." Or run winget install jqlang.jq, then start a new session. Without it, Trove still works when you ask for it. |
| No connection on a company network | Networks that inspect HTTPS with their own certificate are not supported yet. Use Trove through Claude or ChatGPT on the web instead. |
To install without running setup, set the installer arguments first. They apply to that one run only:
$env:TROVE_INSTALL_ARGS = '--no-setup'; irm https://heytrove.ai/install.ps1 | iexIn Git Bash, pass them after the script: curl -fsSL https://heytrove.ai/install.sh | sh -s -- --no-setup. From Git Bash the install skips its questions, such as the tour offer. To get them, run the PowerShell command instead.
Choose which agents get Trove
Setup installs into every detected agent that you have not disabled. Trove supports six agents: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot and Grok. To see or change which agents get Trove, ask your agent:
| Task | What to tell your agent |
|---|---|
| See which agents have Trove | "Show me which of my coding agents have Trove." |
| Turn Trove on for an agent | "Turn on Trove for Codex." |
| Turn Trove off for an agent | "Turn off Trove for Cursor and Grok." |
Your preference is stored in your account and synced to every machine. You can also change it in the web app under Settings ▸ Connected agents.
Claude (app, web, desktop, Cowork)
Claude connects through a custom connector. Open one-tap setup, or add the connector yourself.
- Open Claude Settings → Connectors → Add custom connector.
- Paste
https://heytrove.ai/api/v1/mcpas the server URL. - Sign in to Trove when Claude asks.
Any Claude plan works. On a Team or Enterprise plan, an organization owner can add Trove for everyone in Claude under Organization settings → Connectors.
GitHub Copilot
GitHub Copilot CLI is a coding agent, so it uses the CLI installer above. Setup detects it from the ~/.copilot folder or from the copilot command on your PATH. Its agent token is github-copilot.
Setup installs these files into ~/.copilot (on Windows, %USERPROFILE%\.copilot):
| File | What it does |
|---|---|
skills/trove/SKILL.md | The Trove skill. |
agents/trove-librarian.agent.md, agents/trove-item-reader.agent.md, agents/trove-book-writer.agent.md, agents/trove-book-enricher.agent.md | The four sub-agents. See Librarian and Item reader. |
hooks/trove.json | A session-start hook that adds your project notebook and active shelf. |
copilot-instructions.md (a marked block) | The rules for Plan mode and auto mode. |
If you set COPILOT_HOME, Trove installs into that folder instead. Trove remembers the folder, so background updates find it again when the variable is not set.
The sub-agents use the model of your Copilot session. VS Code Copilot reads the same skills and agents folders, so this install also works there.
Turning Trove off for GitHub Copilot removes only these files and the marked block. Your own skills, agents, hooks and instructions stay.
Tip: Copilot can ask before each trove command that a sub-agent runs. Choose to allow it for the session, or start Copilot with copilot --allow-tool 'shell(trove:*)'. For a headless run, add the same flag to copilot -p.
Grok
Grok Build is a coding agent, so it uses the CLI installer above. Setup detects it from the ~/.grok folder, which the Grok Build installer creates, and installs the Trove skill to ~/.grok/skills/trove/. Grok Build also reads ~/.claude/skills/, but Trove installs its own copy, so it works without Claude Code. Its agent token is grok.
Grok on the web connects through the MCP server instead. Add the server address https://heytrove.ai/api/v1/mcp as a connector in Grok.
ChatGPT
Trove is in ChatGPT's app directory. One click opens it, then sign in.
- Open Trove in ChatGPT and press Connect.
- Sign in to Trove and authorize.
- In a new chat, type @, choose Trove, and ask: "What's in my Trove library?"
On a Business, Enterprise or Edu workspace, an admin may need to allow Trove first.
SSH, containers and CI
Over SSH, in a container or in Codespaces, trove login prints a link and a code. Open the link on your laptop or phone, check the code, and choose Approve.
In CI, set TROVE_API_KEY to a key from API keys. Every trove command uses it and nothing is written to disk. Setup finishes signed out in CI and prints a one-line hint.
TROVE_API_KEY="$TROVE_KEY" trove items listYou can still pipe a key in:
echo "$TROVE_KEY" | trove auth login --with-tokenUpdates
The CLI updates itself in the background. To update now, ask your agent: "Update Trove to the latest version." To see what you have, ask: "Which Trove version do I have?" To turn auto-update off, set TROVE_NO_AUTO_UPDATE=1. Settings live in ~/.trove/config.toml.
Remove Trove
Ask your agent: "Remove Trove from this computer." It asks you to confirm and, in the same message, whether you'd like to tell the team why you're leaving. The question is optional and you can skip it. Then it removes Trove's files from your coding agents, Trove's lines in your shell startup files, ~/.trove and the trove program. Your account, your books and other machines are not changed.
Your API key stays valid. Revoke it in Settings under Developers if you want. If you'd rather tell us later, or Trove is already gone, use the goodbye page.
To turn Trove off in one agent but keep it installed, see "Choose which agents get Trove" above.
Troubleshooting
| Symptom | Fix |
|---|---|
trove: command not found | Open a new terminal, or run source ~/.trove/env.sh (in fish, open a new terminal). |
| Not signed in | Ask your agent to sign you in to Trove, or run trove login. On a remote machine it prints a link to approve. |
| The browser did not open | Wait 60 seconds: trove login then prints a link you can open on any device. Or run trove login --device. |
| The agent never consults Trove | Run trove setup --status, then start a new session. |
| An agent was installed after Trove | The hourly rescan picks it up, or run trove setup. |
| Setup says "No coding agent found" | Use Trove in Claude or ChatGPT instead (see the Claude and ChatGPT sections above), or install a coding agent and run trove setup. |
| GitHub Copilot asks before every trove command | Allow it for the session, or start Copilot with copilot --allow-tool 'shell(trove:*)'. |
| Claude shows two Trove connectors | Remove the custom connector and keep the plugin's own. |
Direct CLI
Your agent runs these through the trove command. You can run them yourself in a terminal, but asking your agent is the recommended way: it picks the right flags, waits for processing and checks the result.
| Task | Command |
|---|---|
| List the agents Trove supports | trove setup --list |
| See which agents have Trove | trove setup --status |
| Turn Trove on for an agent | trove setup --enable codex |
| Turn Trove off for agents | trove setup --disable cursor,grok |
| Update now | trove update |
| Show the installed version | trove version |
| Remove Trove from this computer | trove uninstall |
Agent tokens are claude-code, codex, cursor, gemini-cli, github-copilot and grok. See the CLI reference.