Fixing Docker Module Resolution Error: LangChain Dependencies Investigation

Date: August 25, 2025
Issue: Docker container failing with Cannot find module '@langchain/core/prompts.js' error
Solution: Dependency version alignment across LangChain packages

Problem Description

The dockerized ChatOllama application was experiencing critical module resolution errors during chat operations:

[nuxt] [request error] [unhandled] [500] Cannot find module '/app/.output/server/node_modules/@langchain/core/prompts.js' 
imported from /app/.output/server/chunks/routes/api/models/index.post.mjs

This error occurred consistently across multiple API endpoints (/api/models/chat, /api/instruction, /api/agents) and prevented the application from functioning properly in Docker containers.

Investigation Process

1. Initial Analysis

  • Error Pattern: ESM module resolution failure for @langchain/core/prompts.js
  • Environment: Docker container build process, not local development
  • Affected Files: Server API routes importing from @langchain/core/prompts

2. Container Inspection

Investigation revealed missing export files in the Docker container:

/app/.output/server/node_modules/@langchain/core/prompts.js

/app/.output/server/node_modules/@langchain/core/dist/prompts/index.js

3. Version Conflict Discovery

Found three different versions of @langchain/core in the dependency tree:

  • Project specification: @langchain/core@^0.3.49
  • Actual Docker resolution: @langchain/core@0.3.72 (pulled by deepagents@0.0.1)
  • Legacy versions: @langchain/core@0.1.54 (used by older packages)

The key issue: deepagents@0.0.1 dependency was forcing @langchain/core@0.3.72, while the project specified ^0.3.49, creating version conflicts during Nuxt’s build bundling process.

Root Cause Analysis

Core Issue

Version Mismatch: The newer @langchain/core@0.3.72 has different export structures that weren’t compatible with how Nuxt bundled the modules for Docker deployment.

Why Docker vs Local?

  • Local Development: pnpm’s workspace resolution handled conflicts gracefully
  • Docker Build: Nuxt’s production bundling exposed the version inconsistencies
  • Module Resolution: Different ESM export mappings between versions

Technical Details

// package.json specified
"@langchain/core": "^0.3.49"

// But dependency resolution pulled
"deepagents@0.0.1" → "@langchain/core@0.3.72"

// Resulted in missing exports during bundling

Solution: Dependency Alignment

Approach

Instead of manual file patching, we chose proper dependency management by updating all LangChain packages to compatible versions.

Package Updates Applied

{
  // Core updates for version alignment
  "@langchain/core": "^0.3.49" → "^0.3.72",
  
  // Compatible package updates
  "@langchain/anthropic": "^0.3.19" → "^0.3.26",
  "@langchain/community": "^0.3.41" → "^0.3.53", 
  "@langchain/google-genai": "^0.1.5" → "^0.2.16",
  "@langchain/groq": "^0.0.5" → "^0.2.3",
  "@langchain/ollama": "^0.2.0" → "^0.2.3",
  "@langchain/openai": "^0.5.7" → "^0.6.9",
  
  // Provider-specific updates
  "@langchain/azure-openai": "^0.0.4" → "^0.0.11",
  "@langchain/cohere": "^0.0.6" → "^0.3.4",
  
  // Peer dependency fixes
  "ws": "^8.16.0" → "^8.18.0",
  "zod": "^3.23.8" → "^3.24.1"
}

Implementation Steps

# 2. Reinstall dependencies
pnpm install

pnpm run build

# (Missing parenthesis in server/api/agents/[id].post.ts)

✓ Built in 17.34s

Verification Results

Before Fix

  • Docker Error: Module resolution failure
  • Version Conflicts: 3 different @langchain/core versions
  • Peer Dependencies: Multiple warnings
  • Build Status: Failed in Docker

After Fix

  • Dependency Resolution: All LangChain packages using @langchain/core@0.3.72
  • Local Build: ✅ Successful (pnpm run build)
  • Module Exports: Consistent across all packages
  • Peer Warnings: Reduced to minimal non-critical issues

Best Practices Learned

1. Dependency Management

  • Always align major dependency versions across related package families
  • Use exact or compatible ranges for critical dependencies like LangChain core
  • Regular dependency audits to catch version drift

2. Docker-Specific Considerations

  • Test builds in Docker during development, not just locally
  • Version conflicts manifest differently in containerized builds vs local development
  • ESM module resolution can be sensitive to version mismatches

3. Investigation Approach

  • Container inspection first to understand actual file structure
  • Dependency tree analysis to identify version conflicts
  • Standard tooling over manual fixes for sustainable solutions

Technical Details for Developers

Files Modified

  • package.json: Updated LangChain package versions
  • pnpm-lock.yaml: Regenerated with consistent resolutions
  • server/api/agents/[id].post.ts: Fixed syntax error (missing parenthesis)

Commands for Reproduction

docker exec <container> ls -la /app/.output/server/node_modules/@langchain/core/

docker exec <container> find /app/.output/server/node_modules/@langchain/core -name "*prompt*"

npm list @langchain/core

Prevention Strategy

// package.json - Use stricter version ranges for critical deps
{
  "@langchain/core": "~0.3.72",  // Tilde for patch-level only
  "deepagents": "^0.0.1"         // Ensure compatibility
}

Conclusion

This issue highlights the importance of consistent dependency management in modern JavaScript applications, especially when deploying via Docker. The proper solution involved updating the entire LangChain ecosystem to compatible versions rather than applying manual patches.

Key Takeaways

  1. Version conflicts can manifest differently between local and Docker environments
  2. Dependency alignment is crucial for ESM module resolution
  3. Standard package management is always preferable to manual file fixes
  4. Container-specific testing should be part of the development workflow

The fix ensures ChatOllama’s Docker deployment works reliably while maintaining the standard build process and keeping dependencies up-to-date with the latest LangChain ecosystem improvements.