Power Claude Code with LiteLLM and Together AI

Run Claude Code on Together coding models such as GLM 5.3, Kimi K3, and DeepSeek V4 by routing it through LiteLLM, an open-source gateway you run on your own machine. The LiteLLM proxy serves the Anthropic Messages API used by Claude Code and sends each call to the Together API.

TogetherLink also runs Claude Code on Together models, with no proxy to manage.

Requirements

Claude Code, installed and working.
Python 3.10 or later, with pip or uv.
A Together API key.

Step 1: Install LiteLLM

Install the proxy with pip or uv:

pip install "litellm[proxy]>=1.100.0"

uv tool install "litellm[proxy]>=1.100.0"

Step 2: Export your Together API key

LiteLLM reads the key from TOGETHERAI_API_KEY, not TOGETHER_API_KEY like the Together SDKs. Set it in your terminal before starting:

export TOGETHERAI_API_KEY=<together_api_key>

Step 3: Configure the proxy

Define your models in a config.yaml. Prefix each Together model ID with together_ai/, and give each model a short alias to pass to Claude Code. Any model from the serverless catalog works:

config.yaml

model_list:
  - model_name: glm-5.3
    litellm_params:
      model: together_ai/zai-org/GLM-5.3
      api_key: os.environ/TOGETHERAI_API_KEY
  - model_name: kimi-k3
    litellm_params:
      model: together_ai/moonshotai/Kimi-K3
      api_key: os.environ/TOGETHERAI_API_KEY

Step 4: Start the proxy

Start the server:

litellm --config config.yaml --port 4000 --host 127.0.0.1

The --host 127.0.0.1 flag keeps the proxy reachable only from your own machine.

The proxy takes over this terminal, so open a new one and verify it serves your models:

curl http://localhost:4000/v1/models

You should see your aliases in the response:

{
  "data": [
    { "id": "glm-5.3", "object": "model" },
    { "id": "kimi-k3", "object": "model" }
  ]
}

Step 5: Connect Claude Code

In the same terminal, point Claude Code at the proxy and pass an alias from your config:

export ANTHROPIC_BASE_URL=http://localhost:4000
export ANTHROPIC_AUTH_TOKEN=anything
claude --model glm-5.3

The proxy doesn’t check ANTHROPIC_AUTH_TOKEN, so any value works. Setting it lets Claude Code run without an Anthropic login. If you’re logged in to Claude Code, you can omit it. Claude Code starts as usual, with every request going through the proxy to the Together API.

Claude Code is now running on GLM 5.3, served by Together AI through your own gateway.

Secure the proxy (optional)

A master key makes the proxy require a key on every request. Set one before exposing the proxy beyond your machine, since anyone who can reach an unauthenticated proxy can spend on your Together account. The master key is a secret you make up, not a key issued by Together or LiteLLM:

config.yaml

general_settings:
  master_key: os.environ/LITELLM_MASTER_KEY

Export a random value in the terminal that runs the proxy, then restart it:

export LITELLM_MASTER_KEY=sk-$(openssl rand -hex 16)
litellm --config config.yaml --port 4000 --host 127.0.0.1

Clients now authenticate with the master key. Pass it as ANTHROPIC_AUTH_TOKEN for Claude Code, and as a Bearer token for direct calls to the proxy. To give other people their own keys and budgets instead of sharing the master key, create virtual keys, which also require a Postgres database.

Troubleshooting

Requests fail with No connected db: The proxy has a master key configured and the token you passed doesn’t match it. This usually means LITELLM_MASTER_KEY is unset or has a different value in the terminal where you ran claude or curl. Export the value the proxy started with and try again. The proxy logs Invalid model name passed in model=claude-sonnet-5: Something in your session, such as a subagent pinned to Anthropic’s sonnet alias, requested a model that isn’t in your model_list. Claude Code keeps working, but those requests fail. To route the aliases to a Together model, export ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.3 (and the matching OPUS and HAIKU variables) before starting Claude Code.

Next steps

TogetherLink

Run Claude Code on Together models without hosting a proxy.

Serverless models

Browse every model you can serve through the proxy.

OpenAI compatibility

See how the Together API maps to OpenAI’s format.

Third-party integrations

Use Together through other partner SDKs and gateways.

GUIDES

MODEL QUICKSTARTS

BUILD APPS

BUILD AGENTS

WORK WITH CODING AGENTS

SDK INTEGRATIONS

DEDICATED CONTAINERS

Power Claude Code with LiteLLM and Together AI

Requirements

Step 1: Install LiteLLM

Step 2: Export your Together API key

Step 3: Configure the proxy

Step 4: Start the proxy

Step 5: Connect Claude Code

Secure the proxy (optional)

Troubleshooting

Next steps

TogetherLink

Serverless models

OpenAI compatibility

Third-party integrations

​Requirements

​Step 1: Install LiteLLM

​Step 2: Export your Together API key

​Step 3: Configure the proxy

​Step 4: Start the proxy

​Step 5: Connect Claude Code

​Secure the proxy (optional)

​Troubleshooting

​Next steps

TogetherLink

Serverless models

OpenAI compatibility

Third-party integrations

Requirements

Step 1: Install LiteLLM

Step 2: Export your Together API key

Step 3: Configure the proxy

Step 4: Start the proxy

Step 5: Connect Claude Code

Secure the proxy (optional)

Troubleshooting

Next steps