Connect the Grok API to DSH

Grok and Groq use different setup routes

Grok uses xAI's documentation, console, and API credentials. This site's Groq provider tutorial covers a different service. Keep the two providers' keys separate.

xAI's current Quickstart uses grok-4.7, the API base URL https://api.x.ai/v1, and a Responses example. Older tutorials may use 4.6. Configure a model available to your account and follow the current interface documentation. Official Quickstart

Confirm account and API access

In the xAI console, check your API credentials, the models your team can access, and your billing status. Store a new key safely. Do not publish it in chat logs, article comments, screenshots, or repositories.

Use an official minimal request to confirm that the account can call the target model. Keep the initial setup focused on a short text request. For later work with images, documents, and verification tools, refer to DataCamp's Grok 4.7 tutorial. Official model directory, standalone request in an online tutorial

Complete a short request without DSH

Follow the official Quickstart with a fixed model, address, and protocol. Ask a short question that requires no search, files, or execution tools. Once the request succeeds and returns text, proceed to DSH configuration.

If it fails, record the specific server error. Check whether the credential belongs to the current team, the model is accessible, the request path is correct, or the account has a rate or billing limit. The details in the error response are more useful for diagnosis than knowing that you bought a plan.

If the request works, record non-sensitive details about the model ID, protocol, and response, keeping credentials private. Reuse those established conditions in DSH. Do not change the model merely to make the configuration resemble someone else's screenshot.

Match both protocol and address in DSH

Check the provider directory in your installed version. If an appropriate built-in entry exists, check its protocol and models. Otherwise, inspect the custom model API settings.

The DSH source version used here supports OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages for custom routes. An official Grok Responses example does not belong unchanged in a route configured for a different protocol. DSH protocol implementation

Check each field:

Model discovery is a convenience, not a capability every service guarantees. If discovery fails but generation works independently, check the listing endpoint and response format. Manually adding a confirmed ID may be more useful than repeatedly replacing the key. Official configuration guide

Check an actual DSH task after saving

Select the target model in a new session and request ordinary text first. Then try a read-only tool in an isolated workspace, such as listing files. Check that DSH calls the tool, receives its output, and uses that output correctly in its reply.

Keep two kinds of tools distinct: xAI's server-side tools and DSH's workspace tools. A Python tutorial calling web search or code execution does not establish that those capabilities automatically exist in DSH through the model route.

The useful lesson from DataCamp's 4.7 example is to define acceptance checks before inspecting results. The author tested a Python circuit-review project. Its dependencies and tool calls need separate checks before being used in DSH; this site has not reproduced that integration. The tutorial's verification approach

Optimize costs and long requests afterwards

Once text requests work, consider context, caching, speed, and tool charges. xAI's price table separates models and request conditions. Long-context and other service charges need their own checks; copying a pair of headline rates can miss conditions relevant to your call. Official API pricing

A first connection does not need the longest context. Start with a task small enough to check quickly, then increase scope while retaining a clear request record.

Common questions

Does a missing Grok card mean DSH cannot connect?

That alone is insufficient. Built-in catalogs vary by installed version. A custom route still needs testing with the right protocol and model. This article has not verified your installation.

Why can an official Python example work while DSH fails?

They may use different protocols, models, request fields, or tool declarations. Keep conditions fixed and compare layers rather than changing every setting at once.

Does Grok web search automatically become a DSH tool?

Do not infer that from configuring a model route. Check server-side tools separately from DSH plugins, permissions, and adapter support.

Finish one small task

A saved configuration is only the starting point. Check that the target model completes a task and that you can verify its result. Use the custom provider guide for route settings, and the error center for startup or model failures.

Model setup, selection and cost guides