Tools · Independent guide

DeepSeek Harness Config Generator for rc.7

Build a reviewable settings.yaml provider patch without entering an API key or sending configuration to a server. Choose the official DeepSeek adapter or a custom OpenAI-compatible gateway, then inspect the merged web or headless profile before using it.

Reviewed Aug 19, 2026Developer previewIndependent guide
Evidence status: Official rc.7 provider guide and tagged adapter documentation. Re-check the linked first-party source after a new release candidate.
Track setup progress →

Sharing includes the public link only, without local progress or private configuration.

Browser-only config lab

Build a reviewable provider patch.

Choose an official adapter or a custom OpenAI-compatible gateway. This helper never asks for an API key—only the name of an environment variable that you create outside this site.

LOCALNo uploadNo browser storage
Adapter
Never enter the secret itself. Create the credential through the local DSH UI or your deployment secret manager, then keep only the environment-variable reference here.
settings.yaml patchReview before merging
llm-deepseek:
  apiKeyEnv: "DEEPSEEK_API_KEY"
  thinking: "enabled"
  reasoningEffort: "high"
Inspect the merged profiledsh --profile web --dump-config

Generated syntax is based on the independently reviewed rc.7 adapter documentation. Current releases use a different configuration structure. Do not paste this historical settings.yaml output into a current cordis.patch.yml. Use the current configuration guide instead. This tool does not contact the endpoint, verify a model ID, test image support, or modify any DSH file.

Choose the adapter that matches the endpoint

Use llm-deepseek for the first-party DeepSeek adapter. Use a named llm-pi-ai provider only when the endpoint intentionally implements an OpenAI-compatible protocol and exposes the model identifier you enter.

The generator defaults custom models to text-only. Enabling image input records a capability claim in the patch; it does not probe or certify the endpoint.

  1. Keep provider IDs lowercase and stable because model routes refer to them.
  2. Use HTTPS for remote gateways; plain HTTP is restricted to local loopback endpoints.
  3. Start with one model and add more only after a minimal request succeeds.

Keep credentials outside the generated patch

The form accepts an environment-variable name, never the credential value. It performs all generation in the current page, sends no form request, and writes nothing to browser storage.

Create the real credential in the local DSH settings interface or your deployment secret manager. Do not paste secrets into this site, a task prompt, a plugin configuration, or a public issue.

Validate the merged profile before a task

Copy the patch into the appropriate settings layer, then run the generated --dump-config command. Review the resolved provider, credential reference, model ID and profile before opening a workspace or enabling plugins.

A generated file is not proof of network reachability, protocol compatibility, authentication, model availability or multimodal support. Test those properties with the smallest reversible request.

  1. Compare the resolved tree with the rc.7 config catalog and adapter README.
  2. Stop if the selected profile resolves a different provider or credential source.
  3. Re-check the pinned source after any new release candidate.

First-party sources

Open the pinned or retrieved source before acting on a changed release candidate.

Continue by layer

Configure DeepSeek Harness Models and Providers
Open guide →DeepSeek Harness Custom Provider: Gateway Setup
Open guide →DeepSeek Harness OpenAI-Compatible Endpoint Setup
Open guide →DeepSeek Harness API Key: Setup and Credential Checks
Open guide →DeepSeek Harness Guides: Setup, Providers and Verification
Open guide →DeepSeek Harness Setup Checklist: Save Your Progress
Open guide →