LangGraph Development

Quick reference for running and debugging the research agent.

Quick Commands

Start Development Server

bash

cd /home/rajathdb/research_agent && langgraph dev

Server runs at:

•API: http://127.0.0.1:2024
•Studio UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024
•API Docs: http://127.0.0.1:2024/docs

Run Tests

bash

# All tests
cd /home/rajathdb/research_agent && pytest tests -v

# Specific test file
pytest tests/test_graph/test_nodes.py -v

# Specific test
pytest tests/test_graph/test_nodes.py::test_plan_queries_node -v

# With coverage
pytest tests --cov=research_agent --cov-report=term-missing

Check Server Logs

bash

# If running in background, check output file
tail -f /tmp/claude/-home-rajathdb/tasks/*.output

Environment Setup

Required environment variables in .env:

bash

ANTHROPIC_API_KEY=sk-ant-...
TAVILY_API_KEY=tvly-...

# Optional
LLM_MODEL=claude-sonnet-4-20250514
LLM_TEMPERATURE=0.7

Graph Structure

code

START
  ↓
plan_queries → approval_queries → execute_search → analyze_results
                                        ↑                ↓
                                        └──────── decide_continuation
                                                         ↓
                          ┌──────────────────────────────┴──────────┐
                          ↓                                         ↓
                   plan_queries (loop)                      approval_findings
                                                                    ↓
                                                               synthesize
                                                                    ↓
                                                              write_article
                                                                    ↓
                                                               fact_check ←──┐
                                                                    ↓        │
                                                         ┌──────────┴────────┤
                                                         ↓                   │
                                                   fix_claims ───────────────┘
                                                         ↓
                                                 approval_article
                                                         ↓
                                                     complete
                                                         ↓
                                                        END

Common Errors & Solutions

InvalidUpdateError: Can receive only one value per step

Cause: Multiple nodes updating the same field in parallel (duplicate edges)

Fix:

•Remove duplicate edges from builder.py
•Use Command for conditional routing instead of multiple edges
•For list fields, use Annotated[list[T], add] reducer in state.py

KeyError when accessing state

Cause: Direct state access like state["field"] fails when field is missing

Fix: Always use .get() with defaults:

python

# Bad
topic = state["topic"]

# Good
topic = state.get("topic", "")

Graph stuck at interrupt

Cause: Human-in-the-loop node waiting for input

Fix: In Studio UI, provide the expected approval format:

json

{"approved": true}

Or to reject:

json

{"approved": false, "reason": "Needs more research"}

BlockingError: Synchronous blocking call

Cause: Using synchronous HTTP client in async context

Fix: Use async clients:

python

# Bad
from tavily import TavilyClient
client = TavilyClient()
results = client.search(query)  # Blocks!

# Good
from tavily import AsyncTavilyClient
client = AsyncTavilyClient()
results = await client.search(query)  # Async!

RecursionError in SSL

Cause: Synchronous requests library conflicts with async event loop

Fix: Use httpx or async clients instead of requests

State Fields Reference

Field	Type	Reducer	Description
`topic`	str	-	Research topic
`review_mode`	ReviewMode	-	autonomous/review_before_writing/review_at_each_stage
`max_iterations`	int	-	Max research cycles (default 7)
`sources`	list[Source]	add	Accumulated web sources
`findings`	list[Finding]	add	Extracted research findings
`planned_queries`	list[SearchQuery]	-	Current batch of queries
`article`	str	-	Generated article
`fact_check_results`	list[FactCheckResult]	-	Claim verification results
`current_phase`	ResearchPhase	-	Current execution phase

Testing in Studio

•Start server: langgraph dev
•Open Studio URL in browser
•Create new thread (important: don't reuse corrupted threads!)
•Input format:

json

{
  "topic": "Your research topic here",
  "review_mode": "autonomous",
  "max_iterations": 5
}

Debugging Tips

View graph visualization

Studio UI shows the graph structure. If nodes appear disconnected, check:

•Command[Literal[...]] type annotations on routing nodes
•No duplicate edges from same source node

Check which node is running

Server logs show: langgraph_node=node_name

Inspect state at any point

In Studio, click on a node to see its input/output state

Force restart with clean state

•Stop server (Ctrl+C)
•Start fresh: langgraph dev
•Create NEW thread in Studio (don't resume old ones)