[Bug]: Ollama query API HTTP/500

[mcr-ksh]

Do you need to file an issue?

• I have searched the existing issues and this bug is not already filed.
• I believe this is a legitimate bug, not just a question or feature request.

Describe the bug

ERROR: Ollama chat error: 400: Last message must be from user role
Traceback (most recent call last):
  File "/app/lightrag/api/routers/ollama_api.py", line 498, in chat
    raise HTTPException(
fastapi.exceptions.HTTPException: 400: Last message must be from user role
ERROR: Returning HTTP 500 to client [error_id=01ef498d2914] (HTTPException)

Steps to reproduce

1. Add LightRAG Ollama Endpoint to Open-WebUI.
2. Ask a Question to query the GraphDB
3. If the query goes through at all it will stop at one point
4. Click (>) play/continue button to continue the answer

Expected Behavior

A stable endpoint that will continue the full answer.

LightRAG Config Used

Paste your config here

CHUNK_OVERLAP_SIZE=100
CHUNK_SIZE=1500
CHUNK_TOP_K=30
COSINE_THRESHOLD=0.2
DEFAULT_ENTITY_TYPES = '["Kanzlei", "Person", "Partei", "Unternehmen", "Gericht", "Spruchkörper", "Aktenzeichen", "ECLI", "Rechtszug", "Dokumenttyp", "Rechtsnorm", "Klagegrund", "Verpflichtung", "Berechtigung", "Rechtsfolge", "Gegenstandswert", "Datum", "Ort"]'
EMBEDDING_BATCH_NUM=64
EMBEDDING_BINDING_HOST=http://ai.ironsoftware.de/v1
EMBEDDING_BINDING=openai
EMBEDDING_DIM=4096
EMBEDDING_FUNC_MAX_ASYNC=6
EMBEDDING_MODEL=Qwen/Qwen3-Embedding-8B
EMBEDDING_SEND_DIM=false
EMBEDDING_TIMEOUT=180
EMBEDDING_TOKEN_LIMIT=32768
ENABLE_LLM_CACHE_FOR_EXTRACT=true
ENABLE_LLM_CACHE=true
EXTRACT_OPENAI_LLM_MAX_COMPLETION_TOKENS=4096
EXTRACT_OPENAI_LLM_REASONING_EFFORT=low
FORCE_LLM_SUMMARY_ON_MERGE=5
HOST=0.0.0.0
KG_CHUNK_PICK_METHOD=WEIGHT
KG_CHUNK_PICK_NUMBER=10
LIGHTRAG_DOC_STATUS_STORAGE=RedisDocStatusStorage
LIGHTRAG_GRAPH_STORAGE=Neo4JStorage
LIGHTRAG_KV_STORAGE=RedisKVStorage
LIGHTRAG_VECTOR_STORAGE=QdrantVectorDBStorage
LLM_BINDING_HOST=http://ai.ironsoftware.de/v1
LLM_BINDING=openai
LLM_MODEL=ironai/llm-noreason
LLM_TIMEOUT=3600
LOG_BACKUP_COUNT=5
LOG_LEVEL=DEBUG
LOG_MAX_BYTES=10485760
MAX_ASYNC=6
MAX_ENTITY_TOKENS=15000
MAX_PARALLEL_INSERT=2
MAX_RELATION_TOKENS=25000
MAX_TOTAL_TOKENS=64000
MIN_RERANK_SCORE=0.45
NEO4J_CONNECTION_ACQUISITION_TIMEOUT=30
NEO4J_CONNECTION_TIMEOUT=30
NEO4J_DATABASE=noe4j
NEO4J_KEEP_ALIVE=true
NEO4J_LIVENESS_CHECK_TIMEOUT=30
NEO4J_MAX_CONNECTION_LIFETIME=300
NEO4J_MAX_CONNECTION_POOL_SIZE=100
NEO4J_MAX_TRANSACTION_RETRY_TIME=30
NEO4J_URI=neo4j://irsdb.ironsoftware.de:7687
NEO4J_USERNAME=neo4j
OPENAI_LLM_FREQUENCY_PENALTY=1.0
OPENAI_LLM_MAX_COMPLETION_TOKENS=9000
OPENAI_LLM_MAX_TOKENS=8192
OPENAI_LLM_PRESENCE_PENALTY=1.5
OPENAI_LLM_TEMPERATURE=0.0
OPENAI_LLM_TOP_P=1
PORT=9621
QDRANT_URL=http://irsdb.ironsoftware.de:6333
QUERY_OPENAI_LLM_MAX_COMPLETION_TOKENS=9000
QUERY_OPENAI_LLM_REASONING_EFFORT=medium
QUERY_OPENAI_LLM_TOP_P=0.6
REDIS_CONNECT_TIMEOUT=10
REDIS_MAX_CONNECTIONS=100
REDIS_RETRY_ATTEMPTS=3
REDIS_SOCKET_TIMEOUT=30
REDIS_URI=redis://irsdb.ironsoftware.de:6379
RELATED_CHUNK_NUMBER=10
RERANK_BINDING=cohere
RERANK_BINDING_HOST=http://ai.ironsoftware.de/v1/rerank
RERANK_BY_DEFAULT=True
RERANK_MODEL=Qwen/Qwen3-Reranker-8B
SUMMARY_CONTEXT_SIZE=32000
SUMMARY_LANGUAGE=German
SUMMARY_LENGTH_RECOMMENDED=800
SUMMARY_MAX_TOKENS=2000
TIKTOKEN_CACHE_DIR=./temp/tiktoken
TIMEOUT=1080
TOP_K=60
VERBOSE=True
WORKERS=2

Logs and screenshots

ERROR: Ollama chat error: 400: Last message must be from user role
Traceback (most recent call last):
  File "/app/lightrag/api/routers/ollama_api.py", line 498, in chat
    raise HTTPException(
fastapi.exceptions.HTTPException: 400: Last message must be from user role
ERROR: Returning HTTP 500 to client [error_id=01ef498d2914] (HTTPException)

Additional Information

• LightRAG Version: v1.5.5/0325
• Operating System: Linux/Docker
• Python Version: Linux/Docker
• Related Issues:

[danielaskdd]

Thanks for the report.

LightRAG’s Ollama-compatible API is intended as a convenient way to use LightRAG from clients such as Open WebUI. It is not a stateful chatbot API, and it does not support Open WebUI’s Continue Response action.

Each /api/chat request is treated as an independent RAG query. The final message must be a user message, because it is used as the query for retrieval. Earlier messages may be forwarded as conversation history to the underlying LLM together with the retrieved context, but they do not create retrieval state and are not used to continue a previous generated answer.

Open WebUI’s Continue Response action resends a conversation whose final message has the assistant role. For native Ollama, this can be interpreted as an assistant-prefill/continuation request. That semantic is intentionally not implemented by LightRAG, so such a request is rejected.

Please send a new user message instead, for example: “Please elaborate on …” or ask a new question based on the knowledge base. If possible, disable or avoid the Continue Response action for the LightRAG model in Open WebUI.

The reported error therefore reflects an unsupported request pattern rather than a problem with the underlying LLM or Ollama service.