freellmapi

Point your coding agent at a FreeLLMAPI gateway in one command. The generators read the models your server is actually serving and write the config file each tool expects.

npx freellmapi setup-claude --url http://localhost:3001 --api-key <your-key>

No install step, no account. The unified API key comes from your FreeLLMAPI dashboard (or the tray popover in the desktop app).

Commands

| Command | Tool | | --- | --- | | setup-claude | Claude Code | | setup-codex | Codex CLI | | setup-cline | Cline | | setup-continue | Continue | | setup-aider | Aider | | setup-opencode | OpenCode | | setup-goose | Goose | | setup-qwen | Qwen Code | | setup-roo | Roo Code | | setup-kilo | Kilo Code | | setup-crush | Crush | | setup-dsh | DeepSeek Harness (dsh) | | setup-mimo | MiMo Code (mimo) | | setup-atomcode | AtomCode (atomcode) | | setup-openclaw | OpenClaw | | setup-hermes | Hermes Agent (hermes) | | setup-pi | Pi (pi) | | setup-cursor | Cursor | | setup-generic | Any OpenAI-compatible client | | launch | Run Claude Code with credentials injected into the child process | | launch-codex | Run Codex the same way | | list | Print the supported tools and their base URLs | | keys | Add, list, remove, or test provider keys (keys --help) |

Options

| Flag | Meaning | | --- | --- | | --url URL | Gateway base URL (default http://localhost:3001) | | --api-key KEY | Unified API key | | --profile NAME | Name the generated profile/provider entry | | --model ID | Pin a specific model instead of the catalog default | | --dry-run | Print the diff and write nothing |

FREELLMAPI_URL and FREELLMAPI_API_KEY work in place of --url / --api-key.

Provider keys

The keys commands use a dashboard session token. Set FREELLMAPI_DASHBOARD_TOKEN, or pass --token TOKEN to override it. This is the session returned by dashboard sign-in (stored in the signed-in browser's local storage as freellmapi_dashboard_token). The unified --api-key / FREELLMAPI_API_KEY used by coding agents cannot authenticate these commands. The CLI does not log in or persist the dashboard token.

export FREELLMAPI_URL=http://localhost:3001
export FREELLMAPI_DASHBOARD_TOKEN='<dashboard-session-token>'

npx freellmapi keys add groq                 # hidden provider-key prompt
npx freellmapi keys add groq --key '<key>'   # non-interactive alternative
npx freellmapi keys list
npx freellmapi keys test groq                # check every stored Groq key
npx freellmapi keys test groq --id 7         # check one key
npx freellmapi keys remove groq --id 7

keys list prints each key's ID, platform, enabled chat model count, enabled/disabled state, and stored health status. Counts respect the key's model scope; custom keys count models registered to their endpoint. Credentials and masked keys are never printed. Use the IDs from this list when a platform has multiple keys: remove requires --id in that case, and also checks that the ID belongs to the requested platform. With one matching key, --id is optional.

add saves the key and immediately validates it through the gateway. A failed or inconclusive check exits with code 1 and leaves the key saved for later testing or removal. test also exits with code 1 if any selected key is not healthy or the gateway cannot confirm its validity (for example, after a provider network failure). Successful commands exit with code 0; argument, authentication, and request errors exit with code 1. Validation uses the gateway's usual health checks, including its policy of disabling a key after repeated confirmed failures.

Keyless providers skip the prompt. Custom endpoints need a URL and model registration, so create them in the dashboard; existing custom keys can be listed, tested, and removed here. Requests accept --url / FREELLMAPI_URL and --timeout MS (30,000 ms per request by default). --dry-run is not supported for keys; it is rejected before any request.

Safety

Every generator is non-destructive: it merges into your existing configuration rather than replacing it, and takes a timestamped backup before touching a file that already exists. --dry-run shows the exact diff first.

The two launch commands never write credentials to disk at all — they inject them into the child process environment for that run only.

Requirements

Node.js >= 20.18. A running FreeLLMAPI gateway (install guide).

Links

• Clients & coding agents guide
• Issue tracker

MIT © Tashfeen Ahmed