.env.example

# ---------------------------------------------------------------------------
# Jarvis — example environment. Copy to `.env` and edit:
#
#   cp .env.example .env
#   $EDITOR .env
#   docker compose up -d
#
# This file is the *minimum* config. The full reference lives in
# DEPLOYMENT.md and CLAUDE.md.
# ---------------------------------------------------------------------------

# --- Provider (pick one) ---------------------------------------------------
# Default is `openai`; switch via JARVIS_PROVIDER if you prefer Anthropic /
# Google / Codex / Kimi / Ollama. Each provider has its own _API_KEY.
JARVIS_PROVIDER=openai
OPENAI_API_KEY=sk-replace-me

# Optional model override; per-provider defaults are sane.
# JARVIS_MODEL=gpt-4o-mini

# --- Permission gate -------------------------------------------------------
# `ask` is the safe default — every fs.write / fs.edit / shell.exec call
# requires interactive approval. Switch to `accept-edits` for trusted
# workspaces, `auto` for full automation, `bypass` only if you know what
# you're doing.
JARVIS_PERMISSION_MODE=ask

# --- Tool gates (off by default for safety) --------------------------------
# Uncomment to let the agent edit files / run shell commands inside the
# mounted /workspace volume. These pair with JARVIS_PERMISSION_MODE.
# JARVIS_ENABLE_FS_WRITE=1
# JARVIS_ENABLE_FS_EDIT=1
# JARVIS_ENABLE_FS_PATCH=1
# JARVIS_ENABLE_SHELL_EXEC=1

# --- Auto mode (off by default) --------------------------------------------
# `auto` spawns the background scheduler that picks Approved Requirements
# with an assignee + completed depends_on and runs them unattended.
# JARVIS_WORK_MODE=auto
# JARVIS_WORK_TICK_SECONDS=30
# JARVIS_WORK_MAX_UNITS_PER_TICK=1
# JARVIS_WORK_RUN_TIMEOUT_MS=600000

# --- Logging ---------------------------------------------------------------
# Lift to `debug` when diagnosing tick decisions, approvals, etc.
RUST_LOG=jarvis=info,harness_server=info