products::rolethread :: lite edition

Local-first dataset engineering for narrative AI

RoleThread Lite is a creator-controlled workspace for shaping narrative AI training data: create, validate, repair, organize, merge, and export clean datasets while keeping JSONL files, sidecars, registry data, backups, and workflow decisions local.

RC::V1_SURFACE

Problem

Narrative and conversational training data becomes fragile when edits, tags, character mappings, prompts, and registry metadata drift apart.
Raw JSONL editing makes it too easy to miss malformed entries, duplicate examples, role-order drift, and noisy synthetic patterns.
Fine-tuning preparation needs local ownership, portable metadata, clean export, and recovery paths that stay visible.

Key Features

ChatML dataset creation and management with ShareGPT import/export conversion support.
Validation, conservative repair workflows, character role mapping, and deterministic dataset hardening.
Custom tag categories, alias repair, character registry, system prompts, and portable sidecar metadata.
Protected working copies, merge tools, local backups, optional cloud-backup mirrors, and clean training exports.
Dataset insights for structure, diversity, narrative balance, response length, and metadata integrity.

Philosophy

Local-first by default: datasets, sidecars, registry data, preferences, exports, and backups stay under creator control.
Dataset craftsmanship over magic scoring: validation and repair expose problems before they become training signal.
RoleThread Lite is a dataset workshop, not a chatbot, hosted platform, model trainer, or hidden AI-writing workflow.

release::download

Release path routes through GitHub

Release and download information stays tied to the product repository so source, docs, artifacts, and support context remain close to the code that owns them.

./github_releases

Roadmap

Keep Lite focused on local dataset craftsmanship: creation, validation, metadata, backups, merge, export, and review.
Publish repository-owned documentation through the LatticeFoundry docs surface without duplicating the source of truth.
Preserve the RoleThread namespace for future editions while keeping Studio separate from the Lite release surface.

Docs Boundary

Product repositories remain the documentation source of truth.
LatticeFoundry will render and style product docs without duplicating source content.
Repository-backed rendering will keep the public reading surface aligned with release ownership.