Adaptive Schemas
Structure emerges from your content—no predefined schemas required
Most knowledge systems take one of two approaches:
- Either they require you to define schemas before you can ingest anything—sit down, design your ontology, map out your relationships, then process only content that fits those predetermined structures.
- Or they use simple, predefined schemas that extract basic triplets: subject-predicate-object relationships (think RDF conventions). The latter works well enough for identifying actors and entities, but falls short when you need to understand meaning, capture nuanced arguments, or track how discourse develops over time.
Recurse works quite differently. Structure emerges from your content automatically (we call this adaptive schemas). The system discovers frame patterns in whatever you give it—not just "Person X mentioned Topic Y" but the full semantic structure of how concepts relate, how arguments develop, and how decisions connect.
How It Works
The frames that emerge depend entirely on your content.
Research papers develop frames for Claim, Evidence, and Methodology—slots that capture what's being argued, what supports it, and how it was tested. Bug reports develop frames for Problem, Diagnosis, and Solution. Meeting notes develop frames for Decision, Discussion, and ActionItem. Documentation develops frames for Concept, Procedure, and Parameter.
The system discovers these patterns rather than requiring you to define anything upfront or forcing you into narrow classifications. As you process more content, RAGE evolves the schema registry—recognizing patterns, proposing new frame types, refining existing ones. Process a hundred research papers and the system becomes more accurate at extracting academic argument structures. Process a hundred bug reports and diagnostic patterns become more precisely captured. The schema registry self-governs, bootstrapping itself around whatever content types you provide.
This is what makes Recurse a living substrate—it bootstraps itself by learning from what you add to it.
No more predefined ontologies
Recurse lets you process heterogeneous sources—research papers alongside meeting notes alongside code documentation—without having to manually design ontologies for each type. The schema adapts to whatever patterns exist in your content, creating frames that capture the semantic structure specific to your domain.
| Approach | Schema Definition | Content Handling | Adaptation |
|---|---|---|---|
| Without adaptive schemas | Define upfront for every type | Lost or misclassified if doesn't fit | Requires manual redesign |
| With adaptive schemas | Emerges from content patterns | Automatically discovers new types | Adapts to your domain |
What This Enables
As your knowledge base evolves through temporal versioning, the adaptive schemas preserve the reasoning behind frame type decisions, making the evolution of your knowledge structure itself traceable:
Flexible Ingestion
Upload any type of content—Recurse figures out the structure automatically without requiring type definitions or schema mappings.
Domain Adaptation
The system learns patterns specific to your field through continuous exposure, becoming more accurate over time.
Progressive Refinement
Schema accuracy improves as you process more content in the same domain, refining frame extraction and relationships.
Zero Setup
Start uploading immediately—no taxonomy design, no ontology workshops. Structure emerges from day one.