.env.example

# Copy to .env and fill in. .env is gitignored -- never commit secrets.

# --- Advent of Code ---------------------------------------------------------
# Session cookie, used only to FETCH problems/inputs not already cached under
# years/ (and, if ever enabled, to submit). Get it from adventofcode.com cookies
# after logging in. Secret -- keep it here, not in the repo.
AOC_SESSION=your_session_cookie_here

# Contact string added to the User-Agent, per AoC's automation guidelines.
# Optional; omit to send a generic agent. Do not put PII in the repo -- set it here.
AOC_CONTACT=

# --- Models (local Ollama) --------------------------------------------------
# Where Ollama is listening. Point at a remote host to use a bigger hosted model.
OLLAMA_HOST=http://localhost:11434

# Comma-separated model list the solver may use (must be installed in Ollama).
# Empty = the curated default set (shared/llm/local.py AVAILABLE_MODELS, ordered
# by the measured record and intersected with what's installed) -- the right
# default for most machines. Pin a list here only to force a specific set;
# experiment.py configs that pass models=... override this either way.
SOLVER_MODELS=

# --- Solve behaviour --------------------------------------------------------
# Max repair iterations after a candidate fails verification (default 2).
MAX_REPAIR_ITERATIONS=2

# Optional collaborative-improvement pass with reviewer models (default off;
# adds latency, exercised mainly by integration tests).
ENABLE_COLLABORATIVE_IMPROVEMENT=false

# Actually submit answers to AoC. OFF by default and outward-facing/rate-limited;
# enable deliberately, per problem. (The submission path is currently unwired.)
SUBMIT_SOLUTIONS=false

# --- Misc -------------------------------------------------------------------
LOG_LEVEL=INFO

# Advanced execution/resource limits (timeouts, memory, process caps) are set in
# config/resources.yaml and can be overridden with the matching env vars it
# documents; most users never need to.