Authenticated access to AI providers for Upsun staff.
This downloads the CLI, authenticates via Okta, installs a background service, and configures your shell and AI tools:
curl -fsSL https://ai-gateway.plat.farm/install.sh | bash
This installer avoids SentinelOne antivirus conflicts:
curl -fsSL https://ai-gateway.plat.farm/download.sh | bash
Then run the printed command to complete setup:
~/.local/bin/ai-gateway setup
Run this inside your WSL terminal. It downloads the CLI, authenticates via Okta, and configures your shell and AI tools:
curl -fsSL https://ai-gateway.plat.farm/install.sh | bash
After installation, AI tools work automatically. Restart your terminal (or source your shell config), then run your preferred tool in a project:
claude — Claude Code Recommended (install)codex — OpenAI Codex CLI (install)gemini — Gemini CLI (install)opencode — OpenCode (install)vibe — Mistral Vibe Recommended (install)Your shell also gets the gateway's provider variables, including OPENAI_BASE_URL and the Google/Vertex ones, so your own scripts and anything using the OpenAI or Google SDKs work too — for example make eval. You do not need an API key: the local proxy holds the real credential. For a script built on the Anthropic SDK, run it through ai-gateway run (see the command reference).
On macOS, use ai-gateway run to start the proxy and launch your tool:
ai-gateway run claude — Claude Code Recommended (install)ai-gateway run codex — OpenAI Codex CLI (install)ai-gateway run gemini — Gemini CLI (install)ai-gateway run opencode — OpenCode (install)ai-gateway run vibe — Mistral Vibe Recommended (install)Any command works, not just the tools above, and you do not need an API key — the local proxy holds the real credential:
ai-gateway run make evalai-gateway run python my_script.pyFor a whole shell session instead, keep the proxy running with ai-gateway serve in a dedicated terminal, then set the variables with eval "$(ai-gateway env)" (or ai-gateway env | source under fish) in the terminals you work in. On these platforms nothing starts the proxy in the background, so ai-gateway env on its own leaves the variables pointing at a port with nothing listening. It also leaves out ANTHROPIC_BASE_URL, which only run sets — see the command reference.
WSL typically lacks systemd, so use ai-gateway run to start the proxy and launch your tool:
ai-gateway run claude — Claude Code Recommended (install)ai-gateway run codex — OpenAI Codex CLI (install)ai-gateway run gemini — Gemini CLI (install)ai-gateway run opencode — OpenCode (install)ai-gateway run vibe — Mistral Vibe Recommended (install)Any command works, not just the tools above, and you do not need an API key — the local proxy holds the real credential:
ai-gateway run make evalai-gateway run python my_script.pyFor a whole shell session instead, keep the proxy running with ai-gateway serve in a dedicated terminal, then set the variables with eval "$(ai-gateway env)" (or ai-gateway env | source under fish) in the terminals you work in. On these platforms nothing starts the proxy in the background, so ai-gateway env on its own leaves the variables pointing at a port with nothing listening. It also leaves out ANTHROPIC_BASE_URL, which only run sets — see the command reference.
Commonly used commands:
ai-gateway setup — Complete guided setup (install, login, configure)ai-gateway login — Re-authenticate with the gatewayai-gateway status — Check login and proxy service statusai-gateway disable <tool> — Opt out of gateway for a specific toolai-gateway setup — Complete guided setup (install, login, configure)ai-gateway login — Re-authenticate with the gatewayai-gateway run <command> — Start proxy and run any command (tool or script) with the gateway environmentai-gateway serve — Start proxy in foregroundai-gateway disable <tool> — Opt out of gateway for a specific toolai-gateway setup — Complete guided setup (install, login, configure)ai-gateway login — Re-authenticate with the gatewayai-gateway run <command> — Start proxy and run any command (tool or script) with the gateway environmentai-gateway serve — Start proxy in foregroundai-gateway disable <tool> — Opt out of gateway for a specific tool