Configure a model provider in DeepSeek Harness

DeepSeek Harness separates the agent workflow from the model that powers it. You configure a provider and model in the interface; the harness then sends the context required for the task to that provider.

What to decide first

Before adding a key, check four things in the provider’s own documentation: model availability, API pricing, context limits, and data-retention policy. These details belong to the provider and can change independently of DeepSeek Harness.

Add the provider

Open the model settings in the Web UI, choose the appropriate provider type, add the required endpoint or API key, and select a model. Keep the key in the application’s credential field—not in a prompt or a repository file.

Start with a disposable workspace and a narrow task. Confirm that the selected model responds before adding plugins or granting broader tool access. This isolates model-configuration problems from plugin and workspace-permission problems.

Cost and context controls

Large repositories can create large prompts. Ask the agent to inspect only the relevant directory, exclude generated files, and summarize before making changes. Review the provider dashboard for actual usage and set spending limits there when available.

Privacy boundary

Files included in model context may be transmitted to the configured provider. Do not assume that a local Web UI means local inference. Check the provider endpoint and policy, exclude secrets, and use a local or organization-approved provider when your security requirements demand it.

A reliable first test

Use a small repository and request a read-only explanation. Then approve one small change and inspect the diff. If the model answers but tools fail, the provider is probably configured correctly and the next place to check is workspace permissions or tool approval.