Skip to content

Plugins & Capabilities

IntentWeave ships a lean core that does everything with zero cost. Plugins add capabilities on top — LLM intelligence, graph persistence, or language support. Install only what you need.

LayerWhat You GetCostRequirements
Core (CARI)30+ query modes, architecture visualization, CI drift checks, clone detection, community detection, hub analysis, and more$0Node ≥ 20, Git
+ plugin-llmLLM-generated layer names, selective enrichment (diagram validation, decision tracking, contradiction detection)Pay-per-use (OpenAI)OpenAI API key
+ plugin-kg-liteCypher queries over a local SQLite-backed knowledge graph — no external database needed$0(none)
+ plugin-kgFull Neo4j persistence, semantic queries, impact analysis, cross-document reasoning$0 (self-hosted)Neo4j 5
+ plugin-swiftSwift AST extraction via tree-sitter$0(none)
+ plugin-pythonPython AST extraction via tree-sitter$0(none)

Each row is a use case. Columns show which plugins are needed. Core is always included.

Use CaseCoreplugin-llmplugin-kg-liteplugin-kg
Ranked file retrieval
CI drift detection
Clone detection (exact + structural)
Circular import detection
Unused export detection
Community detection (3 modes)
Architecture layer inference
Interactive HTML report
Focused architecture view
Dependency depth analysis
Boundary violation checks
TODO/rationale inventory
Terminology consistency
Hub / god-node analysis
Cross-group drift
Doc completeness scoring
LLM-generated layer names
LLM-generated directory names
Diagram validation (planned)
Decision tracking (planned)
Contradiction detection (planned)
Config-to-docs sync (planned)
Local Cypher queries
Full semantic KG queries
Impact analysis
Cross-document reasoning
KG persistence
Enrichment + local KG
Enrichment + full KG
Terminal window
iw index build # Core only — $0, < 3 seconds
iw index export --html # architecture report
iw index check --changed $(git diff --name-only HEAD~1)

Plugins needed: none.

Terminal window
iw index build --depth full # deeper annotation matching
iw index export --html --provider openai --model gpt-4o-mini

Plugins needed: plugin-llm (for readable layer/directory names in the report).

Terminal window
iw index build --depth full
iw index export --html --provider openai
iw index clones # find duplicated logic
iw index dep-depth # find deeply nested dependencies
iw index boundary-violations # find cross-package leaks
iw index hubs # find god-nodes

Plugins needed: plugin-llm (optional, for named layers).

Decision archaeology / knowledge management

Section titled “Decision archaeology / knowledge management”
Terminal window
iw run docs/*.md --track open --provider openai -i --persist -v
iw query "What decisions were made about authentication?"
iw impact src/auth/service.ts

Plugins needed: plugin-llm + plugin-kg.

Plugins are auto-discovered from your node_modules. Install them as dev dependencies:

Terminal window
# LLM capability
npm install -D @intentweave/plugin-llm
# Local KG (Cypher over SQLite)
npm install -D @intentweave/plugin-kg-lite
# Full KG (Neo4j)
npm install -D @intentweave/plugin-kg
# Language support
npm install -D @intentweave/plugin-swift
npm install -D @intentweave/plugin-python

Verify what’s loaded:

Terminal window
iw plugin list

No configuration files needed — plugins register their capabilities automatically via the PluginRegistry.

┌─────────────────────────────────────────────────────┐
│ PluginRegistry │
│ │
│ Capabilities: │
│ ┌──────────┐ ┌──────────────┐ ┌───────────────┐ │
│ │ LLM │ │ Persistence │ │ Language │ │
│ │ │ │ │ │ │ │
│ │plugin-llm│ │plugin-kg-lite│ │ plugin-swift │ │
│ │ │ │ plugin-kg │ │ plugin-python │ │
│ └──────────┘ └──────────────┘ └───────────────┘ │
│ │
│ Discovery: scan node_modules for @intentweave/* │
│ Registration: each plugin declares its capability │
│ Resolution: CLI/API asks registry for capability │
└─────────────────────────────────────────────────────┘

Three capability types, each with a clean interface:

  • LLMCapability — text generation, entity extraction, naming
  • PersistenceCapability — store/query knowledge graph triples
  • LanguageCapability — AST extraction for a specific language