config.example.toml

# Example user-level config — copy to ~/.everos/config.toml.
#
#     cp config.example.toml ~/.everos/config.toml
#     $EDITOR ~/.everos/config.toml   # fill in api_key / base_url
#
# Override priority (low → high):
#
#     src/everos/config/default.toml
#       < ~/.everos/config.toml          (THIS FILE)
#       < .env in the working directory
#       < EVEROS_<SECTION>__<KEY> process envs
#       < programmatic init args
#
# The path is configurable: set EVEROS_CONFIG_FILE=/path/to/your.toml to
# point at any location. Missing file = silently skipped (no error).
#
# Comment out anything you do not need — only the fields you set here
# override the shipped defaults; everything else falls back to
# src/everos/config/default.toml.

# ── LLM ───────────────────────────────────────────────
# OpenAI-protocol chat-completions endpoint used by the algo extractors.
[llm]
model    = "gpt-4.1-mini"
api_key  = "sk-..."
base_url = "https://api.openai.com/v1"

# ── Embedding ─────────────────────────────────────────
[embedding]
model    = "Qwen/Qwen3-Embedding-4B"
api_key  = "..."
base_url = "https://api.deepinfra.com/v1/openai"
# Runtime knobs (defaults shown):
timeout_seconds = 30.0
max_retries     = 3
batch_size      = 10
max_concurrent  = 5

# ── Rerank ────────────────────────────────────────────
[rerank]
model    = "Qwen/Qwen3-Reranker-4B"
api_key  = "..."
base_url = "https://api.deepinfra.com/v1/inference"
timeout_seconds = 30.0
max_retries     = 3
batch_size      = 10
max_concurrent  = 5

# ── Other tunables you may want to override here ──────
# [memory]
# root     = "~/.everos"
# timezone = "Asia/Shanghai"
#
# [sqlite]
# busy_timeout_ms = 10000
#
# [lancedb]
# read_consistency_seconds = 5.0