DocsFab helps product and documentation teams turn repository evidence into clear, reviewable product documentation. It can create a first set of docs, update pages affected by a change, review an existing set, or plan a focused restructure. It gathers evidence from the selected source revision and asks when important product intent is missing or sources disagree.
DocsFab is in early access and runs locally through the docsfab command-line
tool. Connect a model through an OpenAI-compatible or Anthropic-compatible
endpoint, including a hosted or self-hosted NVIDIA NIM.
After installing and configuring DocsFab, open a repository and start an inspection-only conversation:
cd /path/to/your-repository
docsfab live-agent --read-only
Then ask: “Review the getting-started docs for clarity and tell me what to improve.” The live agent can answer questions and help with focused edits. In read-only mode it cannot edit files, run write workflows, or run host commands; it can run read-only review workflows. In a normal session, edits are enabled and each host command requires approval in the terminal.
For repeatable work, run a named documentation job:
review-documentation reviews a set of existing pages.create-documentation drafts new pages from selected source material.update-documentation proposes changes for pages affected by a change.restructure-documentation plans and proposes a bounded reorganization.Jobs use selected repository files at a pinned revision. Create and update jobs can also use explicitly bound evidence such as PDF or DOCX files and merge-request or change evidence; supported inputs vary by job. Optional product knowledge can add context when available. Built-in quality checks report their findings separately; their results are not a single universal pass or guarantee. A direct live-agent answer or edit does not by itself run the full job checks or create a workflow run record.
Write jobs can keep a patch for local review or open a draft GitLab merge request. A person reviews and merges the change; delivery is a separate authorized step. Direct live-agent host commands require approval in the terminal. Credentials stay with host-controlled adapters and are not placed in model prompts.
glab and permission to push a branch for workflows that open
a draft GitLab merge request. Read-only reviews and patch-only runs do not
need GitLab push access.NVIDIA technology
Third-party software
glab), for draft merge
request deliveryNVIDIA believes Trustworthy AI is a shared responsibility, and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure the models meet requirements for the relevant industry and use case and address unforeseen product misuse. For more detailed information on ethical considerations for the models, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI concerns here.
DocsFab source code is licensed under the Apache License, Version 2.0. Included skill instructions and documentation are licensed under CC-BY-4.0, and skill scripts are licensed under Apache-2.0. The designated NVIDIA Technical Documentation Style Baseline files are also under CC-BY-4.0. See the repository’s LICENSE and NOTICE files for scope and attribution. Models and endpoints connected to DocsFab are governed by their own terms.
Public installation instructions will be available through the GitHub
repository when the release
is available. If you already have an early-access installation, configure a
model endpoint with docsfab setup, check your setup with docsfab doctor,
and start with the read-only review above.
DocsFab is a documentation agent that reviews, creates, and updates product docs from your source repo, with built-in quality checks.