port: 3500

# All component types (LLMs, vector stores, graph stores, embedders, session stores)
# live under `components:`. The sidecar auto-detects the type by trying each registry
# in order: embedding → session → vector → graph → LLM.
#
# Declaration order matters: declare embedders before vector stores that use them,
# and vector stores before LLM components that reference them via memory_store:.

components:

  # ── Embedder ──────────────────────────────────────────────────────────────
  # Declare before any vector store that needs embeddings.
  # Works with OpenAI and any OpenAI-compatible endpoint (Ollama, LM Studio, etc.).
  # - name: embedder
  #   type: embedding/openai
  #   metadata:
  #     base_url: https://api.openai.com   # or http://localhost:11434/v1 for Ollama
  #     api_key: sk-...                    # or set OPENAI_API_KEY env var
  #     model: text-embedding-3-small
  #     dimensions: "1536"

  # ── Session store (optional) ───────────────────────────────────────────────
  # Defaults to in-memory if omitted (history lost on restart).
  # At most one session store may be configured.
  # - name: session-store
  #   type: session/redis
  #   metadata:
  #     addr: localhost:6379
  #     ttl: "24h"
  #
  # - name: session-store
  #   type: session/postgres
  #   metadata:
  #     dsn: postgres://user:pass@localhost:5432/mydb
  #     table: daimon_sessions
  #     ttl: "24h"

  # ── Vector / Document Stores ───────────────────────────────────────────────
  # Exposed at /v1/memory/{name}/query  etc.
  # The LLM also gets auto-generated search/upsert tools for each store.

  # In-memory BM25 (no external service — dev/testing only):
  # - name: mem
  #   type: inmemory

  # Chroma (handles embeddings server-side):
  # - name: chroma-docs
  #   type: chroma
  #   metadata:
  #     base_url: http://localhost:8000
  #     collection: daimon
  #     create_if_missing: "true"

  # Qdrant (needs embedding_url for semantic search):
  # - name: qdrant-docs
  #   type: qdrant
  #   metadata:
  #     base_url: http://localhost:6333
  #     collection: daimon
  #     create_if_missing: "true"
  #     embedding_url: http://localhost:11434/v1   # Ollama embeddings
  #     dimensions: "768"
  #     # api_key: ...   # Qdrant Cloud

  # Redis Stack (requires redis/redis-stack image for FT commands):
  # - name: redis-docs
  #   type: redis
  #   metadata:
  #     addr: localhost:6379
  #     index: daimon-docs
  #     embedding_url: http://localhost:11434/v1
  #     dimensions: "768"

  # pgvector (PostgreSQL + pgvector extension):
  # - name: pg-docs
  #   type: pgvector
  #   metadata:
  #     dsn: postgres://user:pass@localhost:5432/mydb
  #     table: daimon_documents
  #     embedding_url: http://localhost:11434/v1
  #     dimensions: "768"

  # ── Graph Stores ───────────────────────────────────────────────────────────
  # Exposed at /v1/graph/{name}/cypher  etc.
  # The LLM also gets auto-generated cypher/add_node/add_edge tools for each store.

  # Neo4j (Bolt default; use protocol: "http" for HTTP transport):
  # - name: knowledge-graph
  #   type: neo4j
  #   metadata:
  #     bolt_url: bolt://localhost:7687
  #     protocol: bolt      # "bolt" (default) or "http"
  #     http_url: http://localhost:7474
  #     database: neo4j
  #     username: neo4j
  #     password: secret

  # Memgraph (same Bolt driver as Neo4j; different HTTP endpoint):
  # - name: memgraph-graph
  #   type: memgraph
  #   metadata:
  #     bolt_url: bolt://localhost:7687
  #     protocol: bolt
  #     # username / password optional

  # ── LLM Components ─────────────────────────────────────────────────────────
  # Use memory_store: to enable transparent RAG enrichment from a vector store.
  # The named store must be declared above this component.

  # ── OpenAI ────────────────────────────────────────────────────────────────
  - name: gpt4o
    type: openai
    metadata:
      default_model: gpt-4o
      # api_key: sk-...  # or set OPENAI_API_KEY env var
    # memory_store: chroma-docs   # optional: enable RAG from a vector store
    # models:
    #   o1-preview:
    #     api_key: sk-o1-specific-key
    # defaults:
    #   temperature: 0.7
    #   max_tokens: 2048
    #   top_p: 0.9
    #   stop: ["Human:", "AI:"]
    #   frequency_penalty: 0.0
    #   presence_penalty: 0.0
    #   seed: 42
    #   system: "You are a helpful assistant."

  # ── Anthropic ─────────────────────────────────────────────────────────────
  - name: claude
    type: anthropic
    metadata:
      default_model: claude-opus-4-7
      # api_key: sk-ant-...  # or set ANTHROPIC_API_KEY env var
    # memory_store: chroma-docs   # optional: enable RAG from a vector store
    # models:
    #   claude-haiku-4-5:
    #     api_key: sk-ant-haiku-key
    # defaults:
    #   temperature: 1.0
    #   max_tokens: 4096
    #   top_p: 0.9
    #   top_k: 50          # Anthropic-specific
    #   stop: ["Human:"]
    #   system: "You are a helpful assistant."

  # ── llama.cpp / LM Studio / Ollama (local) ────────────────────────────────
  # - name: local
  #   type: llamacpp
  #   metadata:
  #     base_url: http://localhost:8080/v1   # llama.cpp default
  #     # base_url: http://localhost:11434/v1 # Ollama default
  #     # base_url: http://localhost:1234/v1  # LM Studio default
  #     default_model: llama-3.2-3b-instruct

# MCP servers — daimon connects at startup and injects their tools into every
# chat request automatically. The model can call them; daimon drives the loop.
# mcp_servers:
#   - name: filesystem
#     command: ["npx", "-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
#   - name: github
#     command: ["npx", "-y", "@modelcontextprotocol/server-github"]

telemetry:
  otlp_endpoint: ""  # e.g. "localhost:4318" for a local OTel collector