Issue with js/ts project, files not indexed

[misteral]

Hi @Jakedismo, your project is awesome!!

I work with rust/ts/js project https://github.com/vercel/next.js
I already successfully install codegraph-rust with install-codegraph-osx.sh.
Change path to project and run init & index commands:

❯ codegraph init .
Initializing CodeGraph project...
Path: "."
✓ Created .codegraph/config.toml
✓ Created .codegraph/db/
✓ Created .codegraph/vectors/
✓ Created .codegraph/cache/
✓ Created .gitignore.codegraph

Project initialized successfully!

Next steps:
1. Run 'codegraph index .' to index your codebase
2. Start MCP server: 'codegraph start stdio'
3. Configure Claude Desktop with CodeGraph MCP
4. Experience revolutionary AI codebase intelligence!
❯ codegraph index .
⠁ Indexing project: .
  [00:00:00] [████████████████████████████████████████] 42/42 (100%) 🌳 AST Analysis complete: 5/42 files (✅ 11.9% success) | 📊 7 nodes + 214 edges | ⚡ 0 lines/s | 10,187.7876/s/s | ETA: 0s
⠁ [00:00:00] [                                             ] 0/7 embeddings
                💾 Standard (512 batch) | 0% | 0/s/s | Memory: Optimized | ETA: 0s  
2025-11-05T12:40:15.874503Z ERROR codegraph_vector::embedding: 🚨 CRITICAL: Falling back to random hash-based embeddings
  ✔ Indexing complete
  [00:00:00] [████████████████████████████████████████] 42/42 (100%) 🌳 AST Analysis complete: 5/42 files (✅ 11.9% success) | 📊 7 nodes + 214 edges | ⚡ 0 lines/s | 10,187.7876/s/s | ETA: 0s
  [00:00:00] [█████████████████████████████████████████████] 7/7 embeddings
                💾 💾 Semantic embeddings complete: 7/7 nodes (✅ 100.0% success) | 🤖 default | 📐 384-dim | 🚀 Batch: 512 | 100% | 41,979.0105/s/s | Memory: Optimized | ETA: 0s
  [00:00:00] [████████████████████████████████████████] 7/7 (100%) 📈 Stored nodes & symbols | 11,836.821/s/s | ETA: 0s
  [00:00:00] [████████████████████████████████████████] 214/214 (100%) 🔗 Dependencies resolved: 12/214 relationships (5.6% success) | ⚡ 0.0s | 33,303.0701/s/s | ETA: 0s
🎉 INDEXING COMPLETE!

📊 Performance Summary
┌─────────────────────────────────────────────────┐
│ ⏱️  Total time: 19.01ms                           │
│ ⚡ Throughput: 263.03 files/sec                  │
├─────────────────────────────────────────────────┤
│ 📄 Files:      5 indexed,   37 skipped               │
│ 📝 Lines:    741 processed                          │
│ 💾 Embeddings:      7 generated                      │
└─────────────────────────────────────────────────┘

⚙️  Configuration Summary
Workers: 6 | Batch Size: 512 | Languages:
✅ Excellent embedding success rate (>90%)

🚀 Ready for Revolutionary MCP Intelligence!
Next: Start MCP server with 'codegraph start stdio'

Issue 1: Project has more than 5 files.
Issue 2: ERROR codegraph_vector::embedding: 🚨 CRITICAL: Falling back to random hash-based embeddings

[misteral]

I update project config file .codegraph/config.toml

# CodeGraph Project Configuration
[project]
name = "codegraph-project"
version = "1.0.0"

[indexing]
languages = ["rust", "python", "typescript", "javascript", "go", "java", "cpp"]
exclude_patterns = ["**/node_modules/**", "**/target/**", "**/.git/**", "**/build/**", "**/examples/**"]
include_patterns = ["packages/**/src/**", "lib/**", "**/*.rs", "**/*.py", "**/*.ts", "**/*.js"]
recursive = true
respect_gitignore = true

[mcp]
enable_qwen_integration = true
enable_caching = true
enable_pattern_detection = true

[database]
path = "./.codegraph/db"
cache_path = "./.codegraph/cache"
vectors_path = "./.codegraph/vectors"

[embedding]
provider = "lmstudio"
model = "jina-embeddings-v4-text-code"
lmstudio_url = "http://localhost:1234"
dimension = 1536
batch_size = 64

[llm]
enabled = true  # Set true for local insights
provider = "lmstudio"
model = "qwen/qwen2.5-coder-14b"
lmstudio_url = "http://localhost:1234"
context_window = 64000
temperature = 0.1
insights_mode = "context-only"

[logging]
level = "info"  # Clean TUI output

but nothing was changed. Please help me find my issues.

[Jakedismo]

Could you test embedding with Ollama? I haven't implemented lmstudio embeddings yet as I try to tackle other bugs in the mcp-tools and pipelines first.

[Jakedismo]

Also verify the embedding dimension size of your embedding model vanilla jina-embedding-v4 uses Matryoshka 2048

[misteral]

@Jakedismo Hey Jake, the main issue for now is that files are not being indexed. I think the vector stuff is the next step. Thanks for you time!

[Jakedismo]

Pushed some bug-fixes. Try reinstalling the codegraph with the install-codegraph-clous.sh or osx.sh installer with full features. Ollama embeddings provider with f.ex. all-minilm:latest (fastest for testing) and you need to have the surrealdb instance available (configure your endpoints etc. from config.toml or project level .env as per example). Then try indexing with "codegraph -language typescript -r path/to/project" -r is for recursively going through folders, -l for languages to detect (comma seperated) should respect your .gitignore automatically.
If that resolves the indexing then drop the db and re-index with f.ex. qwen3-embedding:0.6b for better results or 4b or 8b, gemmaembedding also works but is very slow compared to qwen3-family due to the larger corpus it uses. Jina AI cloud embeddings are the current SOTA, relatively fast taking the API calls into account and free up-to 1M tokens.
Tested the http mcp-server and stdio both should work, stdio might still have some logging blocking use. For using the mcp-server remember to configure llm provider also openai, xai, ollama (and maybe lmstudio haven't been focusing on it) should function. Don't have anthropic creds to test that one end to end :D Thx for the input it's hacky still but building it to a better form.