rudraforge-graph — Internal Knowledge Operating System
Identity: OPP-018 (APPROVED) — “the operating memory of RudraForge” Type: Internal tool for RudraForge (not a commercial product) Status: APPROVED, planning complete, implementation not started
What It Is
A project cognition system that ingests the RudraForge repository, represents it as a typed semantic + discourse graph, detects structural/operational/coverage/temporal gaps, and writes accepted corrections back into the repo. It is RudraForge’s self-awareness layer — the system that sees what RudraForge is missing.
Relationship to this wiki: rudraforge-graph is a more sophisticated, code-level graph OS for RudraForge specifically. This wiki is a human-readable compiled knowledge layer for ALL projects. They are complementary — not competing.
Current State
| Metric | Value |
|---|---|
| Status | APPROVED by strategist |
| Planning | Complete (PRODUCT.md, MVP_EXECUTION_PLAN.md) |
| Scaffolding | Next.js app directory structure exists |
| Implementation | Not confirmed started |
| Stack | Next.js + Postgres/SQLite + Cytoscape.js + Graphology |
Architecture
Two graph layers:
- Semantic Graph — typed, durable, derived from repo structure (nodes: products, sessions, skills, decisions, gaps, opportunities)
- Discourse Graph (InfraNodus-style) — emergent from text co-occurrence across Markdown documents
10-step ingestion pipeline:
Parse Markdown/JSON → Segment → Extract entities → Build typed nodes
→ Discourse graph → Compute analytics → Gap candidates
→ Persist → Render → Write-back corrections
File-native: Repository remains source of truth. App indexes and projects. Does NOT replace. Accepted outputs write back into Markdown docs and JSON registries.
Open Gaps
Ingestion Pipeline Not Built (P0)
The 10-step pipeline is fully specified in PRODUCT.md but not implemented. This is the core product.
Graph Persistence Decision Deferred (P1)
MVP uses Postgres or SQLite + Graphology/Cytoscape. Neo4j deferred until graph query complexity justifies it. Decision pending actual implementation.
19 Node Types Defined, 0 Implemented
Full ontology exists in PRODUCT.md (19 node types, 15 edge types). MVP uses 10 node types. None implemented yet.
Key Decisions
| Decision | Chosen | Why |
|---|---|---|
| Graph persistence | Postgres/SQLite first, Neo4j later | Avoid complexity before it’s needed |
| Architecture | File-native (repo = source of truth) | Accepted outputs write back to Markdown |
| Scope | Internal-first (RudraForge only) | Can later expand to client namespaces |
| Stack | Next.js + Cytoscape.js | Consistent with RudraForge product stack |
Relationship to Rudra Wiki
This wiki (what you’re reading) serves a similar function at a higher level: compiled, browsable knowledge across ALL projects. rudraforge-graph serves RudraForge specifically with deeper graph semantics and automated gap detection from code/docs analysis. Long term: rudraforge-graph could feed this wiki automatically.