# ============================================== # OpenZep Configuration # ============================================== # --- LLM (required) --- # Any OpenAI-compatible API: SiliconFlow, OpenAI, local Ollama proxy, etc. # If OpenZep runs in Docker and your gateway/proxy is on the host machine, # use http://host.docker.internal:PORT/v1 instead of localhost / 127.0.0.1. LLM_API_KEY=your-api-key LLM_BASE_URL=https://api.siliconflow.cn/v1 LLM_MODEL=Qwen/Qwen2.5-72B-Instruct LLM_SMALL_MODEL=Qwen/Qwen2.5-7B-Instruct # --- Embedder (optional) --- # If your LLM provider supports embeddings (SiliconFlow, OpenAI, Ollama), # leave these blank and they will reuse LLM_API_KEY / LLM_BASE_URL. # # If you use Anthropic Claude directly (no embedding support), # set a separate embedding provider here. # The same Docker note applies here: prefer host.docker.internal over localhost. # # EMBEDDER_API_KEY= # EMBEDDER_BASE_URL= EMBEDDER_MODEL=BAAI/bge-m3 # --- Graph DB --- GRAPH_DB=neo4j NEO4J_URI=bolt://neo4j:7687 NEO4J_USER=neo4j NEO4J_PASSWORD=password123 # --- SQLite --- SQLITE_PATH=openzep.db # --- Episode ingestion tuning (optional) --- # Slow upstream LLMs can need longer per-episode windows; raise these if you see # "批量发送到Zep失败 ... timed out" (issue #4). Defaults: bulk=90s, single=60s. # GRAPH_BULK_TIMEOUT_SECONDS=90 # GRAPH_SINGLE_TIMEOUT_SECONDS=60 # GRAPH_MAX_CONCURRENT_BATCHES=2 # GRAPH_MAX_BULK_RETRIES=2 # GRAPH_MAX_SINGLE_RETRIES=2 # --- Auth (optional) --- # If set, all API calls must include: Authorization: Bearer # This must EXACTLY match ZEP_API_KEY in your client app (e.g. MiroFish/.env). # Leave empty to disable authentication. # API_KEY=your-openzep-api-key