The one-liner install is the fastest way to wire up token tracking on a single workstation. It's the right path for a pilot, a demo, validating firewall rules, or letting a single engineer self-install. For fleet rollout use MDM instead.

What the one-liner does

The Torii AI Telemetry page shows a one-liner per (tool, OS) combination. On macOS and Linux it's a piped curl into bash; on Windows it's a PowerShell iwr that downloads the installer to %TEMP% and runs it.

Each one-liner:

1. Fetches the install script for that tool + OS from https://assets.toriihq.com/otel/latest/<script-name>.
2. Passes your ingest token to it as a flag. (The Torii endpoint is baked into the script; --endpoint exists only as an override for relays or staging.)
3. Detects the target user's home directory.
4. Makes a timestamped backup of the existing config file - only when the run actually changes it.
5. Writes the telemetry config to the right file for that tool.
6. Exits with a one-line summary of what was written.

The install scripts are idempotent - re-running is safe and produces the same end state, and a re-run that changes nothing skips the write (and the backup) entirely. They touch only the Torii-owned telemetry keys and preserve everything else in the config file (see What the LLM token tracking install scripts do for details).

Before you start

• You have the ingest token from the AI Telemetry page in Torii. (The endpoint, https://ai-events.toriihq.com, is baked into the script - there is nothing else to copy.) The token is shown masked in the page; click Copy on the token row to copy the unmasked value to your clipboard.
• The tool you're configuring (Claude Code, Gemini CLI, Codex CLI, or GitHub Copilot) is installed on the machine - for Copilot that means VS Code with the Copilot Chat extension. The Torii install script writes the telemetry config, but it does not install the CLI itself.
• The machine has outbound HTTPS access to ai-events.toriihq.com and (on Linux only) to assets.toriihq.com.

macOS

Open Terminal on the target machine and paste the one-liner shown on the AI Telemetry page. The exact command depends on which CLI you want to instrument:

# Claude Code
curl -fsSL https://assets.toriihq.com/otel/latest/install-claude-code-telemetry-macos.sh \
  | bash -s -- --token <your-token>

# Gemini CLI
curl -fsSL https://assets.toriihq.com/otel/latest/install-gemini-cli-telemetry-macos.sh \
  | bash -s -- --token <your-token>

# Codex CLI
curl -fsSL https://assets.toriihq.com/otel/latest/install-codex-cli-telemetry-macos.sh \
  | bash -s -- --token <your-token>

# GitHub Copilot (Copilot Chat in VS Code; --email attributes events to the user)
curl -fsSL https://assets.toriihq.com/otel/latest/install-copilot-telemetry-macos.sh \
  | bash -s -- --token <your-token> --email <user@company.com>

What runs on the machine: the macOS installers use native osascript -l JavaScript (JavaScript for Automation, shipped with every macOS since 10.10) to merge JSON safely. No binaries are downloaded at install time. Security reviewers can read the macOS script end-to-end and see exactly what touches disk.

Where the config lands:

• Claude Code → ~/.claude/settings.json
• Gemini CLI → ~/.gemini/settings.json
• Codex CLI → ~/.codex/config.toml
• GitHub Copilot → ~/Library/Application Support/Code/User/settings.json

Linux

The Linux one-liners look identical to macOS - just with -linux.sh in the URL:

# Claude Code
curl -fsSL https://assets.toriihq.com/otel/latest/install-claude-code-telemetry-linux.sh \
  | bash -s -- --token <your-token>

# Gemini CLI
curl -fsSL https://assets.toriihq.com/otel/latest/install-gemini-cli-telemetry-linux.sh \
  | bash -s -- --token <your-token>

# Codex CLI (no helper download - pure awk)
curl -fsSL https://assets.toriihq.com/otel/latest/install-codex-cli-telemetry-linux.sh \
  | bash -s -- --token <your-token>

# GitHub Copilot
curl -fsSL https://assets.toriihq.com/otel/latest/install-copilot-telemetry-linux.sh \
  | bash -s -- --token <your-token> --email <user@company.com>

What runs on the machine: the Claude Code, Gemini CLI, and GitHub Copilot Linux installers download a small statically-linked Go helper (ai-otel-helper, ~1 MB) from assets.toriihq.com/otel/latest/helper/linux-<arch>.gz, verify its SHA256 against a sidecar file, use it to do the JSON merge, and then delete it on exit. This exists because Linux distributions don't ship a built-in JSON parser, and we don't want to require jq / python3 / node on the workstation. The Codex installer uses pure awk on a TOML marker-block and downloads nothing.

Air-gapped Linux fleets: if workstations can't reach assets.toriihq.com directly, mirror the helper internally and set HELPER_BASE_URL before running. See Article 5 for the full procedure.

Where the config lands: same as macOS - ~/.claude/settings.json, ~/.gemini/settings.json, ~/.codex/config.toml.

Windows

Open PowerShell (regular, not elevated, unless you're installing system-wide) and run the two-step one-liner shown on the AI Telemetry page. Example for Claude Code:

iwr -useb 'https://assets.toriihq.com/otel/latest/install-claude-code-telemetry.ps1' `
  -OutFile "$env:TEMP\install-claude-code-telemetry.ps1"
& "$env:TEMP\install-claude-code-telemetry.ps1" -Token '<your-token>'

Substitute install-gemini-cli-telemetry.ps1, install-codex-cli-telemetry.ps1, or install-copilot-telemetry.ps1 in both lines for the other tools. For Copilot, also append -Email '<user@company.com>' so events attribute to the user.

What runs on the machine: the PowerShell installers use native ConvertFrom-Json / ConvertTo-Json. No binaries are downloaded at install time.

Execution policy: if PowerShell refuses to run the script, prepend the second line with an execution policy bypass:

powershell -ExecutionPolicy Bypass -File "$env:TEMP\install-claude-code-telemetry.ps1" -Token '<your-token>'

Where the config lands:

• Claude Code → %USERPROFILE%\.claude\settings.json
• Gemini CLI → %USERPROFILE%\.gemini\settings.json
• Codex CLI → %USERPROFILE%\.codex\config.toml
• GitHub Copilot → %APPDATA%\Code\User\settings.json

Verify it worked

You will see a one-line summary like:

wrote /Users/alice/.claude/settings.json - env (token …a1b2, host=Alice-MBP serial=ABC123 user=alice, log_prompts=0)

To prove end-to-end the telemetry is flowing, open the AI Telemetry page in Torii for the CLI you just configured, then run a test event on the workstation. See Verifying your setup for the exact commands. The expected per-tool test command is:

• Claude Code: echo "hello torii" | claude --print --output-format text
• Gemini CLI: gemini -p "hello torii"
• Codex CLI: codex exec "hello torii"
• GitHub Copilot: open VS Code and send any message to Copilot Chat

The event tail on the AI Telemetry page updates within ~5 seconds.

If you'd rather not pipe a script into bash

The one-liner is just a thin invocation of a published install script. If your environment forbids piping curl into bash, see Manual install for the static JSON / TOML blocks you can write directly into the config files.

Next steps

• Verify your setup - send a test event
• Uninstall token tracking
• Troubleshooting and FAQ