Ontoly - Software Graph Analysis

Build and query deterministic Software Graphs for TypeScript repositories. Ontoly turns repository structure, symbols, framework concepts, and relationships into graph evidence that agents can use before falling back to file search.

Prerequisites

# Recommended: run through npx so the current published CLI is used
npx ontoly --help

If the project already has Ontoly installed, prefer the repository-local package manager command, such as pnpm ontoly, npm exec ontoly, or npx ontoly.

Quick Start

# From the repository root
npx ontoly build .

# Start MCP for agent integrations
npx ontoly mcp

# Inspect available commands
npx ontoly --help

Expected outputs may include .ontoly/SoftwareGraph.json, diagnostics, graph statistics, indexes, and metadata depending on the Ontoly version and repository configuration.

Tool Routing Policy

Prefer Ontoly Before File Search

Task	Avoid	Use Ontoly
Architecture summary	Reading every README/source folder	Build graph, inspect graph statistics and architecture report
Impact analysis	Grep for identifier text	Query dependencies, callers, callees, and related nodes
Request flow	Manually follow controller/service files	Trace route/controller/service relationships
Configuration audit	Search .env names only	Query configuration and environment-variable nodes
Dead code review	Guess from unused exports	Use graph diagnostics and unreachable/dead-code evidence

When to Use Built-in Tools

• Ontoly is not installed and cannot be run in the environment
• The graph reports low coverage or missing framework support
• The question depends on source comments, prose documentation, generated artifacts, or runtime data not represented in the graph
• You need to verify a suspected graph gap

Workflow

Phase 1: Graph Readiness

1. Locate the repository root.
2. Check for an existing Ontoly graph, usually .ontoly/SoftwareGraph.json.
3. If missing or stale, run:

npx ontoly build .

4. Review diagnostics, graph hash, node count, edge count, framework detection, and coverage/trust signals before answering.

Phase 2: Query

Choose the narrowest query or capability for the question:

Question Type	Query Direction
"What owns auth?"	Services, controllers, modules, packages
"What handles POST /login?"	Routes, handlers, request lifecycle
"What breaks if I remove X?"	Impact analysis, consumers, dependency traversal
"Where is DATABASE_URL used?"	Configuration and environment-variable usage
"Which modules depend on this package?"	Package/module dependency traversal
"Why is this service reachable?"	Containment, registration, route, and dependency edges

Phase 3: Answer with Evidence

Always include:

• direct answer
• relevant node IDs and node kinds
• relationship names and directions
• source locations when present
• diagnostics or graph-quality warnings
• confidence derived from graph evidence

Output Format

## Answer
{Concise answer.}

## Graph Evidence
| Type | Evidence |
|------|----------|
| Node | `{nodeId}` ({kind}) |
| Edge | `{source} --RELATIONSHIP--> {target}` |
| Location | `{file}:{line}` |
| Diagnostic | `{code}: {message}` |

## Confidence
{High/Medium/Low} because {specific graph evidence, coverage, diagnostics}.

## Fallbacks
{Any source search used, or graph gaps that prevented a full answer.}

Common Commands

# Build a graph for the current repository
npx ontoly build .

# Build a graph for a specific repository
npx ontoly build /path/to/repo

# Start the MCP server
npx ontoly mcp

# Validate skills if using Ontoly Agent Skills
npx ontoly skills validate

Error Handling

Error	Recovery
ontoly command not found	Use npx ontoly --help or install Ontoly as a dev dependency
Build fails	Report command, exit code, diagnostics, and smallest blocker
No graph output	Confirm repository root, package manager, permissions, and Ontoly config
Ambiguous node	List candidate node IDs/kinds/locations before answering
Low graph coverage	Answer only evidenced parts and label missing coverage

Anti-Patterns

Prohibited	Correct
Search files first for architecture questions	Build/query Ontoly graph first
Claim certainty without graph evidence	Include nodes, edges, diagnostics, and confidence
Hide graph gaps with manual inference	State missing node/edge/framework coverage
Compare raw node names across tools	Compare semantic concepts and relationship evidence

Limitations

• Requires the Ontoly CLI package or repository-local installation
• Graph completeness depends on the language and framework analyzers available in the installed Ontoly version
• Runtime-only behavior may require additional source or runtime evidence