Future Scope

The Internal AI Pipeline (Removed)

Earlier iterations of Docsmith shipped an internal AI pipeline: a laravel/ai SDK-based provider, then a dependency-free OpenAiHttpProvider (OpenAI-compatible chat completions over cURL), a project-level planner that let the model design whole doc sets, a ReviewerAgent quality pass, an agent:run command, and --ai-provider / --ai-model / --ai-api-key / --ai-base-url CLI options. Later it also shipped a structural generate pipeline that produced shallow LLM-free pages and a Playwright media-capture stage.

All of it was removed in favor of the coding-agent architecture: agents bring their own model and keys, and drive Docsmith through MCP (read_source, write_markdown, capture_media, build_site). The media-capture stage was rebuilt properly as the capture_media tool backed by capturist — screenshots and stepped videos of the running app, written straight into the docs source.

The removed code lives in git history (pre-refactor commits on feat/AI, before f808bf1) if anyone ever wants to revisit an unattended/CI pipeline.

Multi-Agent Parallel Generation

For large codebases, doc generation could be parallelized — each module or namespace processed simultaneously. This would dramatically reduce generation time for projects with hundreds of files.

Custom Prompt Templates

Allow users to define documentation style and structure conventions the coding agent should follow (for MCP-driven generation):

Docsmith::serveMcp(
    sourcePath: './app',
    docsSourcePath: './docs-source',
);

RAG from Existing Docs

Feed existing documentation as context for consistency. Newly generated pages would match the tone, terminology, and structure of hand-written docs already in the project.

CI Integration

Run docsmith build as a GitHub Action step to publish the markdown docs as a static site on every push:

- run: docsmith build --source=./docs-source --output=./docs

Web Dashboard

A browser-based UI to monitor, trigger, and review doc generation — showing pipeline progress, media previews, and (when the AI pipeline returns) review scores before publishing.

Plugin System

Third-party agents and tools via Composer:

composer require docsmith/agent-swagger
composer require docsmith/tool-graphviz

Plugins could add new source scanners (for Python, Go, Rust), custom media capturers, or alternative output formats (PDF, API Blueprint).

Multi-Format Export

Beyond HTML sites, generate PDF manuals, OpenAPI specs, or Markdown API references from the same pipeline.

Incremental Generation

Only regenerate docs for changed files since the last build, using git diffs to detect modifications. This would make iteration on large projects near-instant.

Docstring Extraction

Parse PHPDoc, JSDoc, or Python docstrings to enrich generated documentation with inline comments, @param / @return annotations, and type information.