feat: add langgraph js v1

.cursor/plans/migrate-to-local-weaviate-1df91472.plan.md

@@ -0,0 +1,77 @@
+<!-- 1df91472-d73d-47a7-8bf6-a4f0f3c34bca 761d6e0b-bf93-4fb8-acbc-c61e0c924bfb -->
+# Migrate to Local Weaviate with Transformers
+
+## Overview
+
+Replace OpenAI embeddings and Weaviate Cloud with a local Weaviate Docker instance using the `text2vec-transformers` module (`sentence-transformers/multi-qa-MiniLM-L6-cos-v1`) for cost-free embeddings.
+
+## Key Changes
+
+### 1. Update Docker Infrastructure
+
+**File: `docker-compose.yml`**
+
+Add Weaviate service with text2vec-transformers module:
+
+- Weaviate core service on port 8080
+- text2vec-transformers service with `sentence-transformers-multi-qa-MiniLM-L6-cos-v1` model
+- CPU-only configuration (ENABLE_CUDA: 0)
+- Persistent volume for data storage
+
+### 2. Update Weaviate Connection Logic
+
+**File: `backend/ingest.py`**
+
+Changes:
+
+- Replace `weaviate.connect_to_weaviate_cloud()` with `weaviate.connect_to_local()`
+- Remove `WEAVIATE_API_KEY` usage (local instance doesn't need auth)
+- Configure WeaviateVectorStore to use Weaviate's built-in vectorizer instead of external embeddings
+- The `text2vec-transformers` module will handle vectorization automatically
+
+**File: `backend/retrieval.py`**
+
+Changes:
+
+- Update `make_weaviate_retriever()` to connect to local instance
+- Modify `make_text_encoder()` to support `weaviate/text2vec-transformers` option
+- Handle embedding model configuration for local vs cloud scenarios
+
+### 3. Update Embedding Configuration
+
+**File: `backend/embeddings.py`**
+
+Update `get_embeddings_model()` to return `None` or a placeholder when using Weaviate's built-in vectorizer, since Weaviate handles embeddings internally.
+
+**File: `backend/configuration.py`**
+
+Update default `embedding_model` from `"openai/text-embedding-3-small"` to `"weaviate/text2vec-transformers"` to reflect the new local setup.
+
+### 4. Environment Variables
+
+Update `.env` file:
+
+- Change `WEAVIATE_URL` from cloud URL to `http://localhost:8080`
+- `WEAVIATE_API_KEY` can be removed or left empty (not needed for local)
+
+### 5. Optional: Update Index Names
+
+**File: `backend/constants.py`**
+
+Consider renaming index names to reflect the new embedding model (e.g., replace `OpenAI_text_embedding_3_small` suffix with `transformers_multi_qa_MiniLM`).
+
+## Technical Notes
+
+- Weaviate's text2vec-transformers module vectorizes text automatically at ingestion and query time
+- No external API calls for embeddings = zero cost
+- CPU-only transformer inference will be slower than GPU but functional
+- First run will download the transformer model (~80MB)
+- Data persists in Docker volume between restarts
+
+### To-dos
+
+- [ ] Add Weaviate and text2vec-transformers services to docker-compose.yml with proper configuration
+- [ ] Modify backend/ingest.py to connect to local Weaviate and remove external embedding dependency
+- [ ] Modify backend/retrieval.py to support local Weaviate connection and built-in vectorizer
+- [ ] Update backend/embeddings.py and backend/configuration.py for Weaviate built-in vectorizer
+- [ ] Start Docker services and test ingestion with local Weaviate to verify embeddings work correctly

.github/workflows/eval-js.yml

@@ -0,0 +1,57 @@
+name: Eval JS
+
+on:
+  workflow_dispatch:
+
+concurrency:
+  group: eval-js-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  run_eval:
+    runs-on: ubuntu-latest
+    environment: evals
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "20"
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          version: 10.23.0
+          run_install: false
+
+      - name: Get pnpm store directory
+        shell: bash
+        run: |
+          echo "STORE_PATH=$(pnpm store path --silent)" >> $GITHUB_ENV
+
+      - name: Setup pnpm cache
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.STORE_PATH }}
+          key: ${{ runner.os }}-pnpm-store-${{ hashFiles('backend-js/pnpm-lock.yaml') }}
+          restore-keys: |
+            ${{ runner.os }}-pnpm-store-
+
+      - name: Install dependencies
+        working-directory: backend-js
+        run: pnpm install --frozen-lockfile
+
+      - name: Run E2E tests
+        working-directory: backend-js
+        env:
+          LANGSMITH_API_KEY: ${{ secrets.LANGSMITH_API_KEY }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          GROQ_API_KEY: ${{ secrets.GROQ_API_KEY }}
+          OLLAMA_API_KEY: ${{ secrets.OLLAMA_API_KEY }}
+          OLLAMA_BASE_URL: ${{ vars.OLLAMA_BASE_URL }}
+          WEAVIATE_URL: ${{ vars.WEAVIATE_URL }}
+          WEAVIATE_GRPC_URL: ${{ vars.WEAVIATE_GRPC_URL }}
+          WEAVIATE_API_KEY: ${{ secrets.WEAVIATE_API_KEY }}
+          RECORD_MANAGER_DB_URL: ${{ secrets.RECORD_MANAGER_DB_URL }}
+        run: pnpm test:e2e

.github/workflows/eval.yml

@@ -10,7 +10,7 @@ concurrency:
 jobs:
   run_eval:
     runs-on: ubuntu-latest
-    environment: Evaluation
+    environment: evals
     steps:
       - uses: actions/checkout@v4
 
@@ -48,7 +48,11 @@ jobs:
         env:
           LANGSMITH_API_KEY: ${{ secrets.LANGSMITH_API_KEY }}
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-          WEAVIATE_URL: ${{ secrets.WEAVIATE_URL }}
+          GROQ_API_KEY: ${{ secrets.GROQ_API_KEY }}
+          OLLAMA_API_KEY: ${{ secrets.OLLAMA_API_KEY }}
+          OLLAMA_BASE_URL: ${{ vars.OLLAMA_BASE_URL }}
+          WEAVIATE_URL: ${{ vars.WEAVIATE_URL }}
+          WEAVIATE_GRPC_URL: ${{ vars.WEAVIATE_GRPC_URL }}
           WEAVIATE_API_KEY: ${{ secrets.WEAVIATE_API_KEY }}
+          RECORD_MANAGER_DB_URL: ${{ secrets.RECORD_MANAGER_DB_URL }}
         run: uv run pytest backend/tests/evals

_scripts/clear_index.py

@@ -2,52 +2,130 @@
 
 import logging
 import os
+from dotenv import load_dotenv
+
+# Load environment variables from .env file
+load_dotenv()
 
-import weaviate
-from langchain.embeddings import OpenAIEmbeddings
 from langchain.indexes import SQLRecordManager, index
-from langchain.vectorstores import Weaviate
+from langchain_weaviate import WeaviateVectorStore
+from backend.embeddings import get_embeddings_model
+from backend.constants import (
+    OLLAMA_BASE_EMBEDDING_DOCS_URL,
+    OLLAMA_BASE_URL,
+    WEAVIATE_GENERAL_GUIDES_AND_TUTORIALS_INDEX_NAME,
+)
+from backend.utils import get_weaviate_client
 
+logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
-WEAVIATE_URL = os.environ["WEAVIATE_URL"]
-WEAVIATE_API_KEY = os.environ["WEAVIATE_API_KEY"]
+RECORD_MANAGER_DB_URL = os.environ.get(
+    "RECORD_MANAGER_DB_URL",
+)
+
 RECORD_MANAGER_DB_URL = os.environ["RECORD_MANAGER_DB_URL"]
-WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_OpenAI_text_embedding_3_small"
+WEAVIATE_URL = os.environ.get("WEAVIATE_URL")
+WEAVIATE_GRPC_URL = os.environ.get("WEAVIATE_GRPC_URL")
+
+WEAVIATE_API_KEY = os.environ.get("WEAVIATE_API_KEY")
 
 
 def clear():
-    client = weaviate.Client(
-        url=WEAVIATE_URL,
-        auth_client_secret=weaviate.AuthApiKey(api_key=WEAVIATE_API_KEY),
-    )
-    vectorstore = Weaviate(
-        client=client,
-        index_name=WEAVIATE_DOCS_INDEX_NAME,
-        text_key="text",
-        embedding=OpenAIEmbeddings(),
-        by_text=False,
-        attributes=["source", "title"],
-    )
-
-    record_manager = SQLRecordManager(
-        f"weaviate/{WEAVIATE_DOCS_INDEX_NAME}", db_url=RECORD_MANAGER_DB_URL
-    )
-    record_manager.create_schema()
-
-    indexing_stats = index(
-        [],
-        record_manager,
-        vectorstore,
-        cleanup="full",
-        source_id_key="source",
-    )
-
-    logger.info("Indexing stats: ", indexing_stats)
-    logger.info(
-        "LangChain now has this many vectors: ",
-        client.query.aggregate(WEAVIATE_DOCS_INDEX_NAME).with_meta_count().do(),
-    )
+    embedding = get_embeddings_model(base_url=OLLAMA_BASE_URL)
+
+    with get_weaviate_client(
+        weaviate_url=WEAVIATE_URL,
+        weaviate_grpc_url=WEAVIATE_GRPC_URL,
+        weaviate_api_key=WEAVIATE_API_KEY,
+    ) as weaviate_client:
+        collection_name = WEAVIATE_GENERAL_GUIDES_AND_TUTORIALS_INDEX_NAME
+
+        # First, directly delete all documents from Weaviate collection
+        # This ensures we delete everything, not just what's tracked in record manager
+        try:
+            collection = weaviate_client.collections.get(collection_name)
+
+            # Get count before deletion
+            initial_count = collection.aggregate.over_all().total_count
+            logger.info(
+                f"Found {initial_count} documents in collection before deletion"
+            )
+
+            if initial_count > 0:
+                # Fetch all object UUIDs and delete them individually
+                # This is the most reliable way to delete all documents
+                import weaviate.classes.query as wq
+
+                deleted_count = 0
+                batch_size = 100
+
+                while True:
+                    # Fetch a batch of objects (only get UUIDs, not full data)
+                    objects = collection.query.fetch_objects(limit=batch_size)
+
+                    if not objects.objects:
+                        break
+
+                    # Delete each object individually
+                    for obj in objects.objects:
+                        try:
+                            collection.data.delete_by_id(obj.uuid)
+                            deleted_count += 1
+                        except Exception as e:
+                            logger.warning(f"Failed to delete object {obj.uuid}: {e}")
+
+                    logger.info(
+                        f"Deleted batch of {len(objects.objects)} documents (total: {deleted_count})"
+                    )
+
+                    # If we got fewer objects than batch_size, we're done
+                    if len(objects.objects) < batch_size:
+                        break
+
+                logger.info(
+                    f"Successfully deleted {deleted_count} documents directly from Weaviate collection: {collection_name}"
+                )
+            else:
+                logger.info("Collection is already empty")
+
+        except Exception as e:
+            logger.warning(f"Could not delete directly from collection: {e}")
+            logger.info("Falling back to record manager cleanup...")
+
+        vectorstore = WeaviateVectorStore(
+            client=weaviate_client,
+            index_name=collection_name,
+            text_key="text",
+            embedding=embedding,
+            attributes=["source", "title"],
+        )
+
+        record_manager = SQLRecordManager(
+            f"weaviate/{collection_name}",
+            db_url=RECORD_MANAGER_DB_URL,
+        )
+
+        record_manager.create_schema()
+
+        # Also clean up record manager to keep it in sync
+        indexing_stats = index(
+            [],
+            record_manager,
+            vectorstore,
+            cleanup="full",
+            source_id_key="source",
+        )
+
+        logger.info(f"Indexing stats: {indexing_stats}")
+        num_vecs = (
+            weaviate_client.collections.get(collection_name)
+            .aggregate.over_all()
+            .total_count
+        )
+        logger.info(
+            f"General Guides and Tutorials now has this many vectors: {num_vecs}"
+        )
 
 
 if __name__ == "__main__":

_scripts/evaluate_chains.py

@@ -29,7 +29,7 @@
     "openai": "gpt-3.5-turbo-1106",
     "anthropic": "claude-2",
 }
-WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_OpenAI_text_embedding_3_small"
+WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_transformers_multi_qa_MiniLM"
 
 
 def create_chain(
@@ -112,12 +112,11 @@ def create_chain(
 
 
 def _get_retriever():
-    WEAVIATE_URL = os.environ["WEAVIATE_URL"]
-    WEAVIATE_API_KEY = os.environ["WEAVIATE_API_KEY"]
+    WEAVIATE_URL = os.environ.get("WEAVIATE_URL", "http://localhost:8080")
 
+    # Note: This uses the old Weaviate client API v3. Consider upgrading to v4.
     client = weaviate.Client(
         url=WEAVIATE_URL,
-        auth_client_secret=weaviate.AuthApiKey(api_key=WEAVIATE_API_KEY),
     )
     weaviate_client = Weaviate(
         client=client,

_scripts/evaluate_chains_agent.py

@@ -20,15 +20,14 @@
 from langsmith.evaluation.evaluator import EvaluationResult
 from langsmith.schemas import Example, Run
 
-WEAVIATE_URL = os.environ["WEAVIATE_URL"]
-WEAVIATE_API_KEY = os.environ["WEAVIATE_API_KEY"]
-WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_OpenAI_text_embedding_3_small"
+WEAVIATE_URL = os.environ.get("WEAVIATE_URL", "http://localhost:8080")
+WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_transformers_multi_qa_MiniLM"
 
 
 def search(inp: str, callbacks=None) -> list:
+    # Note: This uses the old Weaviate client API v3. Consider upgrading to v4.
     client = weaviate.Client(
         url=WEAVIATE_URL,
-        auth_client_secret=weaviate.AuthApiKey(api_key=WEAVIATE_API_KEY),
     )
     weaviate_client = Weaviate(
         client=client,

_scripts/evaluate_chains_improved_chain.py

@@ -30,7 +30,7 @@
     "openai": "gpt-3.5-turbo-1106",
     "anthropic": "claude-2",
 }
-WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_OpenAI_text_embedding_3_small"
+WEAVIATE_DOCS_INDEX_NAME = "LangChain_Combined_Docs_transformers_multi_qa_MiniLM"
 
 
 def search(search_queries, retriever: BaseRetriever):
@@ -148,12 +148,11 @@ def create_chain(
 
 
 def _get_retriever():
-    WEAVIATE_URL = os.environ["WEAVIATE_URL"]
-    WEAVIATE_API_KEY = os.environ["WEAVIATE_API_KEY"]
+    WEAVIATE_URL = os.environ.get("WEAVIATE_URL", "http://localhost:8080")
 
+    # Note: This uses the old Weaviate client API v3. Consider upgrading to v4.
     client = weaviate.Client(
         url=WEAVIATE_URL,
-        auth_client_secret=weaviate.AuthApiKey(api_key=WEAVIATE_API_KEY),
     )
     weaviate_client = Weaviate(
         client=client,