DocsFab Support Matrix
Use this matrix to prepare an environment for the current DocsFab development build used by internal testers. Follow Install DocsFab for the installation commands. This page records platform coverage and the scope of model testing; it does not designate a default model.
Last reviewed: October 14, 2026.
Hardware
DocsFab runs locally and calls the model endpoint you configure. Hosted models do not require a local GPU, CUDA, or a container runtime. If you run a local NVIDIA NIM, its GPU, driver, and container requirements apply to the model server. Refer to Use a Local NIM.
The following resources are required on the DocsFab host:
Software
Operating System
Platform coverage depends on the operation you run:
Linux and macOS installations must also satisfy the operating-system requirements of Node.js 24.21.0 or later. The tested configuration below records a specific host; it is not a minimum OS-version guarantee for every distribution.
Runtime Dependencies
Prepare the dependencies required by your intended operation:
The current source installation does not require npm authentication to the DocsFab package registry. Packaged development installations require separate registry access and npm authentication, as described in Install a Development Build.
Verified Configurations
The following retained qualification establishes installation and command isolation on a specific configuration:
This configuration establishes those installed-product checks. It does not establish a passing full WSL regression suite or successful workflows with every model. Model results are listed separately below.
Tested Models
These models have recorded DocsFab workflow tests on the specified routes. The results describe the tested revision and task, rather than every possible combination of model, endpoint, platform, and workflow.
The eight-case matrix covers create, review, update, and restructure through both the CLI and the live agent. The following table summarizes retained results:
The internal gateway tests require access to that service. They do not establish
results for the same model on the direct OpenAI API, direct Anthropic API,
native Azure or Bedrock APIs, or a local NIM container. A model ID containing
azure/ or aws/ identifies the model served by the tested gateway.
Other models can be configured when their endpoint supports the selected API, streaming, and native tool calling. A successful connection check or an installed composition does not establish workflow quality. For the meaning of a complete workflow result, refer to Evaluating DocsFab.
Network Access
Prepare access to the services used by your chosen operation:
External API Service Access
Choose credentials for the service that actually hosts your model:
- For OpenAI or Anthropic, use an API key from that provider’s product console.
- For OpenAI-compatible or Anthropic-compatible, obtain the service’s model ID, base endpoint URL, API format, and key. Compatibility depends on the endpoint’s API; selecting a compatible provider does not add native cloud authentication or discover an endpoint for you.
- For a local NIM, follow its deployment requirements and the local connection instructions.
Store model credentials through docsfab setup or the approved host secret
facility. Do not put keys in repository files or command-line arguments.
For the setup prompts, refer to Configure the Model.
Requirements by Operation
For model, repository, and delivery prerequisites by operation, follow Check Readiness.
Pre-Installation Checklist
Confirm the prerequisites that apply to your intended operation:
- The host platform and Node.js version meet the requirements above.
- You can clone DocsFab and write to its checkout and your npm global prefix.
- You have the model ID, service key, and endpoint details needed for model-backed work.
- Linux command isolation is available when the workflow requires validation checks.
- GitLab authentication and publication permissions are available when you need draft merge request delivery.