CCL runs Claude Code against a different model provider — DeepSeek, GLM, a local model, whatever — without you editing config or exporting environment variables by hand.

It works because Claude Code only speaks the Anthropic Messages API. Any provider that also speaks that API will work with Claude Code; the model just needs to be reachable at the right URL with the right key. CCL knows those URLs and keys, sets them, and launches claude.

ccl --provider deepseek does three things:

1. Reads the deepseek entry from your config — base URL, API key, model name.
2. Sets the ANTHROPIC_* environment variables to match, and removes any pre-existing ANTHROPIC_* / CLAUDE_CODE_* variables from the environment so they can't leak through.
3. Replaces itself with claude via syscall.Exec.

There's no proxy or translation layer in the request path — CCL exits the moment claude starts, and the process you're left talking to is Claude Code, configured. (Transformer rules for non-Anthropic-shaped APIs exist but are opt-in; nothing runs between you and the model unless you set them up.)

go install github.com/dotcommander/cclauncher/cmd/ccl@latest

~/go/bin must be on your PATH. From source:

git clone https://github.com/dotcommander/cclauncher
cd cclauncher
just install

ccl # pick a provider; pre-selects the default
ccl --provider deepseek # this session only; skips the picker
ccl --provider synthetic "fix the null pointer in main.go"

Arguments after the provider flag pass through to claude unchanged.

Note on -p: CCL uses -p for --provider; Claude Code uses -p for print mode. The first -p/--provider goes to CCL, everything after passes through to claude.

ccl -p deepseek -p "fix this bug" # provider=deepseek, claude in print mode
ccl -p "fix this bug" # error: "fix this bug" is not a provider

Use the long form --provider to avoid ambiguity.

For the full docs index, see docs/README.md.

CCL ships with eleven providers in three categories. They're pre-configured; you only supply the key.

Cloud — synthetic, deepseek, minimax, zai, openrouter, wafer. Each reads one environment variable:

export DEEPSEEK_API_KEY="sk-..."
ccl --provider deepseek
export OPENROUTER_API_KEY="sk-or-..."
ccl --provider openrouter

Local — llamabarn, lmstudio, llamacpp, omlx. No key; the model server runs on your machine and CCL points at localhost:

ccl --provider lmstudio # LM Studio, localhost:1234
ccl --provider llamacpp # llama.cpp, localhost:8080
ccl --provider llamabarn # LlamaBarn, localhost:2276

Anthropic — claude. Auth is handled by the claude CLI's own OAuth; nothing to set.

Endpoints, environment variables, and per-provider notes: docs/providers.md.

On first run, CCL writes ~/.config/cclauncher/config.yaml with all providers filled in. You supply the API key through an environment variable; CCL can't guess it.

Run bare ccl at an interactive terminal to open the provider picker. It pre-selects the configured default, so press Enter to launch it or pick another provider for that launch only. In pipes or CI, bare ccl uses the configured default without prompting.

Set that default by editing cli.defaultProvider in ~/.config/cclauncher/config.yaml.

Override a single key without editing the config — CCL_<PROVIDER>_API_KEY takes precedence:

export CCL_DEEPSEEK_API_KEY="sk-..."

If you select a provider whose key is missing, CCL stops before launching and tells you which variable to set, rather than producing a 401 mid-session.

Config schema and interpolation rules: docs/configuration.md. Local model setup: docs/local-models.md.

just build # build ./ccl
just install # build + symlink to ~/go/bin/ccl
just test # run tests
just lint # golangci-lint
just dev # go run, no build

Providers are defined in YAML, not Go. To add one, update internal/config/default-config.yaml, then run go run ./internal/tools/gen-config-examples so committed examples stay generated from the canonical default config. CCL sets provider env vars generically. See docs/providers.md.

MIT