Build Your Own Module

If the built-in modules don’t fit your data, you can build your own forge module. A module extends the base pipeline with domain-specific scanners, templates, and review criteria.

Module structure

A custom forge module provides three things:

Scanners — pull raw data from your domain sources
Templates — structured formats for the kinds of beliefs your domain produces
Review criteria — domain-specific rules for what makes a belief valid

What each piece does

Scanners

Pull raw data from your domain sources. Receive CLI config (paths, URLs, credentials). Return documents that the extract stage will process.

Examples:

Read files from a directory
Pull records from an API
Query a database
Scrape a website

Templates

Structured formats for the kinds of knowledge your domain produces. Templates help the extract stage produce consistent, well-structured beliefs.

Review criteria

Domain-specific criteria for the review stage. What makes a belief valid in your domain? What should the reviewer look for?

Examples:

“Claims must cite a specific version number”
“Measurements must include units and conditions”
“Recommendations must reference at least one observed failure”

Current approach

Until the plugin API is available, the recommended approach is to use the document forge with domain-specific prompts:

# Use the document forge for any text-based source
reasonsforge document --sources-dir ./my-data --domain "My domain context"

The document forge handles chunking, summarization, belief extraction, derivation, review, and repair for any text input. The --domain flag provides context that shapes the LLM prompts.

For more control, you can run individual pipeline steps:

reasonsforge forge chunk-docs --input-dir ./my-data
reasonsforge forge summarize --input-dir ./my-data
reasonsforge forge propose-beliefs --all
reasonsforge forge accept-beliefs
reasonsforge forge derive-review-repair --domain "My domain context"