Skip to content

providers.md

Latest commit

 

History

History
218 lines (158 loc) · 6.56 KB

providers.md

File metadata and controls

218 lines (158 loc) · 6.56 KB
Raw

Providers

CCL ships with 11 pre-configured providers across three categories. Each provider maps to a specific model and API endpoint. CCL sets the correct ANTHROPIC_* environment variables and calls claude — no proxy or translation layer runs unless you configure transformer rules.

ccl providers

Use the command above to see the providers and models loaded from your current ~/.config/cclauncher/config.yaml. The tables below describe the committed defaults in internal/config/default-config.yaml; examples/config.yaml.example is generated from that same canonical file.

Cloud (Native APIs)

These providers use their own API endpoints and require separate keys.

synthetic

export SYNTHETIC_API_KEY="sk-..."
ccl --provider synthetic
Field Value
Endpoint https://api.synthetic.new/anthropic
Model hf:zai-org/GLM-4.7
Small/Fast hf:zai-org/GLM-4.7
Env var SYNTHETIC_API_KEY

synthetic routes through Synthetic.new to access GLM-4.7. Swap the model for any HuggingFace slug Synthetic hosts by editing ~/.config/cclauncher/config.yaml.

deepseek

export DEEPSEEK_API_KEY="sk-..."
ccl --provider deepseek
Field Value
Endpoint https://api.deepseek.com/anthropic
Model deepseek-v4-pro
Small/Fast deepseek-v4-flash
Env var DEEPSEEK_API_KEY

minimax

export MINIMAX_API_KEY="sk-..."
ccl --provider minimax
Field Value
Endpoint https://api.minimax.io/anthropic
Model MiniMax-M2.7
Small/Fast MiniMax-M2.7-highspeed
Env var MINIMAX_API_KEY

zai

export ZAI_API_KEY="sk-..."
ccl --provider zai
Field Value
Endpoint https://api.z.ai/api/anthropic
Model glm-4.7 (override: ZAI_MODEL)
Small/Fast glm-4.5-air (override: ZAI_SMALL_MODEL)
Env var ZAI_API_KEY

zai supports model overrides without editing the config file:

export ZAI_MODEL="glm-4.7"
export ZAI_SMALL_MODEL="glm-4.5-air"

openrouter

export OPENROUTER_API_KEY="sk-or-..."
ccl --provider openrouter
Field Value
Endpoint https://openrouter.ai/api
Model deepseek/deepseek-v3.2
Small/Fast deepseek/deepseek-v3.2
Env var OPENROUTER_API_KEY

OpenRouter exposes an Anthropic Messages-compatible route at /api (its "Anthropic Skin"). Swap the model to any OpenRouter slug — e.g. anthropic/claude-sonnet-4.5, moonshotai/kimi-k2.5 — by editing the model/smallFastModel in ~/.config/cclauncher/config.yaml.

wafer

export WAFER_API_KEY="..."
ccl --provider wafer
# or shorthand
ccl -p wafer
Field Value
Endpoint https://pass.wafer.ai
Model GLM-5.1 (override: WAFER_MODEL)
Small/Fast GLM-5.1 (override: WAFER_SMALL_MODEL)
Env var WAFER_API_KEY

Wafer Pass routes to frontier open models via an Anthropic-compatible endpoint. Available models include GLM-5.1, DeepSeek-V4-Pro, and Qwen3.5-397B-A17B (model names are case-insensitive). Override without editing config:

export WAFER_MODEL="DeepSeek-V4-Pro"
export WAFER_SMALL_MODEL="Qwen3.5-397B-A17B"

Anthropic (OAuth)

claude

ccl --provider claude
Field Value
Endpoint https://api.anthropic.com
Auth claude CLI's own OAuth flow — CCL doesn't manage credentials

No API key needed. If you haven't authenticated, claude will prompt you to log in.

Local

Local providers connect to a server running on your machine. No API key is required by default, and no internet traffic is sent for the LLM call itself.

See local-models.md for server setup instructions and model recommendations.

ccl --provider llamabarn   # localhost:2276
ccl --provider lmstudio    # localhost:1234
ccl --provider llamacpp    # localhost:8080
ccl --provider omlx        # localhost:8000

llamabarn

Field Value
Endpoint http://localhost:2276 (override: LLAMABARN_BASE_URL)
Model local (override: LLAMABARN_MODEL)
Auth LLAMABARN_API_KEY (optional)

lmstudio

Field Value
Endpoint http://localhost:1234 (override: LMSTUDIO_BASE_URL)
Model openai/gpt-oss-20b (override: LMSTUDIO_MODEL)
Auth Any non-empty token — defaults to lmstudio

Tip: LM Studio validates that the Authorization header is present but doesn't check the token value. The default lmstudio token works with any loaded model.

llamacpp

Field Value
Endpoint http://localhost:8080 (override: LLAMACPP_BASE_URL)
Model local (override: LLAMACPP_MODEL)
Auth Any non-empty token — defaults to llamacpp

llama-server supports the Anthropic Messages API natively when started with --port 8080. CCL connects directly without a translation layer.

omlx

Field Value
Endpoint http://localhost:8000 (override: OMLX_BASE_URL)
Model Qwen3.5-9B-MLX-4bit (override: OMLX_MODEL)
Small/Fast gemma-4-e2b-it-4bit (override: OMLX_SMALL_MODEL)
Auth OMLX_API_KEY (optional)

omlx targets a local MLX-backed Anthropic-compatible server on macOS. Override the endpoint and models without editing config:

export OMLX_BASE_URL="http://localhost:8000"
export OMLX_MODEL="Qwen3.5-9B-MLX-4bit"
export OMLX_SMALL_MODEL="gemma-4-e2b-it-4bit"

Switching Providers

Switch for a single session with --provider:

ccl --provider deepseek "review this PR"

Run bare ccl at an interactive terminal to open the provider picker. It pre-selects cli.defaultProvider; press Enter to launch it or pick another provider for that launch only.

Set the default provider by editing cli.defaultProvider in ~/.config/cclauncher/config.yaml. When stdin is not a terminal, bare ccl uses that default without prompting.

Environment Variable Overrides

Every provider supports a CCL_<PROVIDER>_API_KEY override that takes precedence over the config file:

export CCL_SYNTHETIC_API_KEY="sk-other-key"   # overrides SYNTHETIC_API_KEY for synthetic
export CCL_DEEPSEEK_API_KEY="sk-other-key"    # overrides DEEPSEEK_API_KEY for deepseek

Full auth priority: CCL_<PROVIDER>_API_KEY > config authToken (with ${VAR} interpolation) > built-in default.

See configuration.md for the full interpolation reference.