Modulus – Cross-Repo Knowledge Orchestration for AI Agents

TL;DR

TL;DR: Modulus builds a unified knowledge graph across your repositories, giving AI coding agents full cross-file context that individual RAG pipelines miss.

Source and Accuracy Notes

• Official site: https://modulus.so
• Launch HN: https://news.ycombinator.com/item?id=43058504

What Is Modulus?

Modulus is a developer tool that connects multiple code repositories into a shared knowledge graph. Unlike single-repo RAG setups, it tracks symbol references, call chains, and semantic relationships across repos — so when an AI agent asks “how does auth work here,” it can trace the actual flow across your entire org’s codebase.

It is designed for teams where AI coding agents need to operate across multiple services or microservices, not just a single project context.

Setup Workflow

Step 1: Connect Your Repositories

# Install the Modulus CLI
npm install -g @modulus/cli

# Authenticate
modulus auth login

# Add your first repository
modulus repo add my-org/service-a
modulus repo add my-org/service-b

Step 2: Index and Build the Knowledge Graph

# Trigger a full index across all connected repos
modulus index --all

# Check index status
modulus status

Step 3: Query Across Repos

# Ask a cross-repo question via CLI
modulus query "How does the auth token get validated across services?"

# Or use the REST API
curl -X POST https://api.modulus.so/v1/query \
  -H "Authorization: Bearer $MODULUS_KEY" \
  -d '{"question": "Trace the payment flow from checkout to settlement"}'

Deeper Analysis

How the Knowledge Graph Works

Modulus parses your code using tree-sitter based parsers, extracting:

• Symbol tables (functions, classes, interfaces)
• Cross-reference edges (who calls whom, who imports what)
• Semantic labels (auth, payment, user, etc.)

These are stored in a graph database (Nebula Graph under the hood) and queried at inference time to construct a context window that includes only the truly relevant code paths.

vs. Single-Repo RAG

Traditional RAG chunks code by file or by chunk, losing cross-file relationships. Modulus preserves the graph. For a large monorepo or multi-service org, this means the difference between an agent asking “where is the auth middleware?” and it actually knowing the full call chain from login() to validateToken() to refreshSession() across repos.

Current Limitations

• Only supports TypeScript/JavaScript and Python currently
• Self-hosted option requires a Nebula Graph instance
• The knowledge graph is not real-time — re-indexing is needed after major refactors

Practical Evaluation Checklist

• Does the tool handle monorepos with 10+ packages?
• Is the index queryable via both CLI and API?
• Does it preserve cross-repo symbol resolution?
• Is the pricing predictable for small teams?
• Does it integrate with existing tools (Cursor, Copilot, etc.)?

Security Notes

• API keys are scoped to read-only queries by default
• Code never leaves your infrastructure in the self-hosted deployment
• No telemetry on private code in the cloud tier

FAQ

Q: Does Modulus work with private repositories? A: Yes. Both cloud and self-hosted modes support private GitHub/GitLab repos with standard OAuth credentials.

Q: How often does the knowledge graph update? A: On-demand via CLI or webhook-triggered on push events. Real-time streaming is on the roadmap.

Q: What happens if a symbol is renamed across repos? A: Modulus tracks rename refactors during indexing and updates edges accordingly, keeping the graph consistent.

Conclusion

Modulus solves the context fragmentation problem that plagues AI coding agents in multi-repo environments. If your team is already running into “the agent doesn’t understand our codebase” barriers, a knowledge graph approach like Modulus is worth evaluating. The CLI-first experience makes it easy to pilot before committing to a self-hosted deployment.