Launch Any Coding Agent from the CLI
The LLM Gateway CLI can now start any supported coding agent pre-wired to the gateway — Claude Code, OpenCode, Codex CLI, DevPass Code, and eight more. One command, one API key, 200+ models, and every request tracked in your dashboard.
Every coding agent has its own idea of how to talk to a gateway: one wants env vars, another an auth.json, a third a TOML file, a fourth per-session flags. Wiring each one to LLM Gateway meant hunting through its docs first. llmgateway launch replaces all of that — one command that starts any supported coding agent with LLM Gateway configured out of the box.
One command, any agent
1# Interactive agent picker2npx @llmgateway/cli launch3
4# Launch a specific agent (shortcuts work too: `llmgateway claude`)5npx @llmgateway/cli launch claude6npx @llmgateway/cli launch opencode7
8# Pick a model — launcher flags go before the agent name9npx @llmgateway/cli launch -m gpt-5.5 claude10
11# Everything after the agent name is passed to the agent itself12npx @llmgateway/cli launch claude --continue13
14# See all supported agents and which are installed15npx @llmgateway/cli launch --list16
17# Inspect what would run without launching18npx @llmgateway/cli launch --dry-run codex1# Interactive agent picker2npx @llmgateway/cli launch3
4# Launch a specific agent (shortcuts work too: `llmgateway claude`)5npx @llmgateway/cli launch claude6npx @llmgateway/cli launch opencode7
8# Pick a model — launcher flags go before the agent name9npx @llmgateway/cli launch -m gpt-5.5 claude10
11# Everything after the agent name is passed to the agent itself12npx @llmgateway/cli launch claude --continue13
14# See all supported agents and which are installed15npx @llmgateway/cli launch --list16
17# Inspect what would run without launching18npx @llmgateway/cli launch --dry-run codexTwelve agents are supported at launch, each configured the way its own tooling expects:
| Agent | Launch | Configuration |
|---|---|---|
| DevPass Code | llmgateway devpass-code | First-party agent — key refreshed in its auth.json |
| Claude Code | llmgateway claude | ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN env vars |
| OpenCode | llmgateway opencode | Built-in llmgateway provider |
| Empryo | llmgateway empryo | empryo --set-key llmgateway |
| SoulForge | llmgateway soulforge | soulforge --set-key llmgateway |
| Codex CLI | llmgateway codex | Per-session -c overrides (no config file changes) |
| Autohand Code | llmgateway autohand | OPENAI_BASE_URL + OPENAI_API_KEY env vars |
| Pi | llmgateway pi | Provider added to ~/.pi/agent/models.json |
| Kimi Code | llmgateway kimi | Provider added to ~/.kimi-code/config.toml |
| MiMo Code | llmgateway mimo | Provider routed through the gateway in mimocode.json |
| OpenClaw | llmgateway openclaw | Provider added to ~/.openclaw/openclaw.json |
| Hermes Agent | llmgateway hermes | hermes setup run with gateway values on first launch |
Whichever agent you pick, requests flow through your LLM Gateway key — so model routing, fallback, cost tracking, and per-agent usage attribution all work exactly as they do everywhere else. If an agent isn't installed, the launcher prints its official install command instead of failing cryptically.
Keys are verified before anything starts
The launcher resolves your API key from --key, the LLMGATEWAY_API_KEY environment variable, or the key stored by llmgateway auth login --key — in that order — and verifies it against the gateway before starting the agent. A stale key (say, one you rolled last week) is reported with its exact source, and the launcher falls back to the next valid one, prompting for a fresh key only if none works. No more launching an agent just to watch its first request 401.
Works with any LLM Gateway API key on every plan — PAYG credits and DevPass alike.