Choose your model.
Give it a real job.

Start with what you have.

Ducoro brings your Bots together in one place. To start with your first Bot, choose a model route below.

Bring your subscription.

If you already use Claude Code or Codex with a supported subscription, start there. The tool owns its sign-in and provider access.

1. Install your chosen tool and sign in.

   Follow the provider’s setup instructions and confirm you can send a message from that tool.

   Claude Code setupCodex setup
2. Open Ducoro and bring in your Bot.

   Choose the tool you signed in to. Ask Ducoro to verify that it is ready before handing over work.

3. Use the model and limits of your plan.

   The model choices and usage available to that Bot come from the tool and your provider’s subscription.

Connect your model server.

Your Bots can use a model hosted on a GPU server you control. Ducoro connects over your network; model inference runs on that server. Choose hardware, memory, and quantization for the model you plan to run.

This is a configuration example. Replace the address with your server’s reachable endpoint and use the exact model ID it reports.

A dedicated AI system such as NVIDIA DGX Spark can serve models that fit its memory. DeepSeek V4.1 Flash requires a server or cluster sized for its full model weights, context, and serving workload. Select the model and hardware together.

1. Prepare your inference server.

   Follow the model’s deployment guide. Use a serving engine and model configuration that support the tool calls your work needs.

2. Make it reachable from Ducoro.

   Enable API-key authentication on the model service. Check its address and model ID, and keep the service running and reachable on your network.

3. Ask Ducoro to connect and verify.

   Configure the endpoint, model ID, and API key in Ducoro. Verify a model call and a file task before handing over recurring work.

Prefer a hosted model? DeepSeek’s API serves V4.1 Flash as deepseek-flash at https://api.deepseek.com, using your own API account. Hosted inference sends model requests to the provider. Connected tools use their own services and permissions.

Hand over a weekly feedback review.

Choose a folder with a week of customer feedback, the previous week’s brief, and your current priorities. Keep customer names or IDs consistent so repeated comments can be counted correctly.

Tell your Bot where to find the folder and what to make.

Your Bot may need permission to read or write in that folder. Confirm the folder and the intended result before it starts.

Look at what came back.

Open the result and compare it with the notes. A completed message is the beginning of your review.

• weekly-brief.md is saved in the folder you chose, and the source files are intact.
• The customer counts and quotes match the original feedback, including repeated comments from the same customer.
• Each proposed priority has sources and reasons. Missing information and conflicting evidence are visible.
• Your Bot brings the priorities back to you for a decision.

Once you have checked the brief and its sources, discuss how the review should continue. Agree on a schedule, the feedback folder to follow, and which decisions should come back to you. Keep Ducoro, the model server, and the required tools available for that arrangement.