LiteLLM on Seedrise

LiteLLM settings at a glance

Checked against LiteLLM's documentation on 2026-11-02.

Why put Seedrise behind LiteLLM

Setup in five minutes

1. Get a Seedrise key. Request a rate card, receive your console login, and create a key for your LiteLLM deployment in the console, with the models it may call and a monthly budget.
2. Store the keys. Export your Seedrise key as SEEDRISE_API_KEY on the proxy host, next to LITELLM_MASTER_KEY for LiteLLM itself. LiteLLM requires the master key to start with sk-.
   bashCopy

   export SEEDRISE_API_KEY="sk-..."      # your Seedrise key
   export LITELLM_MASTER_KEY="sk-..."    # admin key for your proxy

3. Add the models to config.yaml. Each Seedrise model is openai/ plus its ID, with api_base set to https://api.seedrise.ai/v1. model_name is the name your apps will call.
   yamlCopy

   model_list:
     - model_name: claude-sonnet-5-5
       litellm_params:
         model: openai/claude-sonnet-5-5
         api_base: https://api.seedrise.ai/v1
         api_key: os.environ/SEEDRISE_API_KEY
         timeout: 600
     - model_name: gpt-6.1-sol
       litellm_params:
         model: openai/gpt-6.1-sol
         api_base: https://api.seedrise.ai/v1
         api_key: os.environ/SEEDRISE_API_KEY
         timeout: 600
     - model_name: deepseek-v3.2
       litellm_params:
         model: openai/deepseek-v3.2
         api_base: https://api.seedrise.ai/v1
         api_key: os.environ/SEEDRISE_API_KEY
     - model_name: kimi-k2.5
       litellm_params:
         model: openai/kimi-k2.5
         api_base: https://api.seedrise.ai/v1
         api_key: os.environ/SEEDRISE_API_KEY
   
   general_settings:
     master_key: os.environ/LITELLM_MASTER_KEY

4. Start the proxy. Run litellm --config config.yaml. The examples here use port 4000, as LiteLLM's own documentation does.
   bashCopy

   litellm --config config.yaml

5. Send a test request. Call http://localhost:4000/v1/chat/completions with your master key and a model_name from the config.
   bashCopy

   curl http://localhost:4000/v1/chat/completions \
     -H "Authorization: Bearer $LITELLM_MASTER_KEY" \
     -H "Content-Type: application/json" \
     -d '{"model":"claude-sonnet-5-5","messages":[{"role":"user","content":"Hello"}]}'

Python SDK, without the proxy

The same three settings work in litellm.completion() directly.

import os
import litellm

response = litellm.completion(
    model="openai/claude-sonnet-5-5",
    api_base="https://api.seedrise.ai/v1",
    api_key=os.environ["SEEDRISE_API_KEY"],
    messages=[{"role": "user", "content": "Summarise this changelog in three bullets."}],
)
print(response.choices[0].message.content)

Models to start with

Context and output limits as at 1 November 2026. Every model on Seedrise, with its list price, is in the model library.

Claude through LiteLLM's Anthropic route

If you rely on Claude features that live in the Anthropic Messages format, such as cache_control prompt caching, route Claude with the anthropic/ prefix and the root URL. LiteLLM appends /v1/messages itself.

model_list:
  - model_name: claude-sonnet-5-5
    litellm_params:
      model: anthropic/claude-sonnet-5-5
      api_base: https://api.seedrise.ai
      api_key: os.environ/SEEDRISE_API_KEY
      timeout: 600

• Use one route per model name. Your apps keep calling claude-sonnet-5-5 either way.
• Keep the openai/ prefix for GPT, DeepSeek, Kimi and the other models on your key.

Things to know

• api_base ends in /v1 for openai/ models. LiteLLM adds /chat/completions itself, so do not paste the full path.
• The openai/ prefix tells LiteLLM which client to use and is not sent upstream. Seedrise receives the bare ID, such as claude-sonnet-5-5.
• Seedrise does not serve the Responses API (/v1/responses) yet. LiteLLM forwards /v1/responses calls for openai/ models straight upstream, so they return 404. If a client of your proxy needs that endpoint, declare the model with the custom_openai/ prefix instead; LiteLLM's documentation says it bridges those calls through Chat Completions.
• Reasoning models can stream for several minutes. Keep timeout under litellm_params at 600 seconds or more for them.
• For spend tracking, add input_cost_per_token, output_cost_per_token and cache_read_input_token_cost under model_info with the rates on your rate card, so LiteLLM's spend logs match your invoice.

Check it is working

• GET /v1/models on the proxy lists your model_name entries; on Seedrise it lists the IDs your key can call.
• Usage appears in the Seedrise console within a minute, by key and model.

# models your proxy exposes
curl http://localhost:4000/v1/models -H "Authorization: Bearer $LITELLM_MASTER_KEY"

# models your Seedrise key can call
curl https://api.seedrise.ai/v1/models -H "Authorization: Bearer $SEEDRISE_API_KEY"

Known errors

Still stuck? Email support@seedrise.ai with the request ID from the response headers, the time and the model ID.

For teams

Run one proxy for the team and keep Seedrise keys on the server, not on laptops.

• Give developers LiteLLM virtual keys with their own budgets (they need LiteLLM's database). The Seedrise key behind the proxy never leaves the server.
• Use one Seedrise key per team or environment, each in its own environment variable, when you want spend split by team on the Seedrise invoice.
• Set the Seedrise key's monthly budget as a hard ceiling above the total of your LiteLLM budgets.
• Add the proxy's outbound IP to the key's IP allow-list, so a leaked Seedrise key does not work anywhere else.
• To rotate, create a new key in the console, update the environment variable, restart the proxy, then disable the old key.

LiteLLM questions