Integration guide

DeepSeek Harness on Seedrise.

Add Seedrise as a custom model service and run Claude Opus, Sonnet and Haiku 5.5 inside DeepSeek Harness at wholesale rates, with Australian regions when you need them.

Get a rate card →Jump to setup

Why run DeepSeek Harness through Seedrise

Claude inside your harness

DeepSeek Harness speaks Anthropic Messages, so Claude models work natively: tools, streaming and long context, no translation layer.

Wholesale rates, one invoice

Below list prices for every Claude model on the rate card, stepping down with volume, billed in USD or AUD.

Keys you control

Model allow-list, budget, IP allow-list and expiry per key. Usage by project in the console and on the invoice.

Setup in five minutes

Get your Seedrise key. Request a rate card, receive your console login and create a key allowed to call the Claude models.
Add a custom model service. In DeepSeek Harness settings, add a custom model service and choose the Anthropic Messages protocol.
Enter the connection. Base URL https://api.seedrise.ai (no /v1), API key = your Seedrise key, model claude-sonnet-5-5 or any Claude ID on your rate card.
Pick the model in the harness. Select the new service as the active model. Add Opus or Haiku as separate entries if you switch between them.
Run a task. The first response confirms the route; the request shows in the Seedrise console under your key.

Check it is working

The first task completes and the Seedrise console shows a request for claude-sonnet-5-5 (or the model you picked).
Tool calls and streaming behave exactly as against Anthropic directly.
A 404 on count_tokens, if the harness calls it, is expected and harmless.

For teams

One key per team keeps spend separated on the invoice.
Restrict keys to the Claude models you have approved; the harness will only see those.
Ask for Australian regions on the key if your data must stay in Sydney or Melbourne.

Troubleshooting and questions

Should I choose OpenAI or Anthropic protocol? +

Anthropic Messages, with the root URL https://api.seedrise.ai. The OpenAI-compatible endpoint also works, but the Anthropic protocol keeps Claude features native.

The service shows 401 or 403. +

401: the key is wrong, disabled or expired. 403: the model is not on the key's allow-list, or your IP is outside its allow-list. Fix it in the console.

Can I use non-Claude models here? +

Through the OpenAI-compatible endpoint (https://api.seedrise.ai/v1, Chat Completions protocol) you can call every model on the rate card.

Where does my data go? +

Through the Seedrise gateway to the model's region. Nothing is stored by Seedrise and nothing is used for training. See security and data.

Ready to switch DeepSeek Harness to wholesale rates?

Per-model rates in writing within one business day, then a console login and a trial key.

Get a rate card →

Also: full docs · model library · security and data