.env.example

# ============================================================
# Nora — Environment Configuration
# ============================================================
# Copy this file to .env and replace all <REPLACE_...> values.
#
# Generate secrets with:
#   node -e "console.log(require('crypto').randomBytes(32).toString('hex'))"
# setup.sh keeps this file at mode 0600; setup.ps1 removes inherited access and
# grants only the current user, SYSTEM, and local Administrators on Windows.
# Production Compose and Helm mount core values under /run/secrets. Nora's
# production entrypoint also accepts the conventional *_FILE variables below
# for custom orchestrators; a non-empty direct value takes precedence.
# setup materializes Compose-only copies under this gitignored owner-only
# directory so read-only containers can mount them without inline secrets.
NORA_COMPOSE_SECRETS_DIR=.secrets/compose
# ============================================================

# ── Required ─────────────────────────────────────────────────
JWT_SECRET=<REPLACE_WITH_RANDOM_SECRET>            # min 32 chars — used to sign JWTs
ENCRYPTION_KEY=<REPLACE_WITH_64_HEX_CHARS>         # 32-byte hex key for AES-256-GCM (run the command above)
NORA_BACKUP_ENCRYPTION_KEY=<REPLACE_WITH_64_HEX_CHARS>  # 32-byte hex key for managed backup archives; do not rotate casually
NORA_AGENT_HUB_API_KEY_HASH_SECRET=<REPLACE_WITH_RANDOM_SECRET>  # setup/update auto-generates when missing; do not rotate casually
NORA_API_KEY_HASH_SECRET=<REPLACE_WITH_RANDOM_SECRET>   # primary workspace/API token hash secret; setup preserves the legacy fallback value
# JWT_SECRET_FILE=/run/secrets/JWT_SECRET
# ENCRYPTION_KEY_FILE=/run/secrets/ENCRYPTION_KEY
# NORA_BACKUP_ENCRYPTION_KEY_FILE=/run/secrets/NORA_BACKUP_ENCRYPTION_KEY
# NORA_AGENT_HUB_API_KEY_HASH_SECRET_FILE=/run/secrets/NORA_AGENT_HUB_API_KEY_HASH_SECRET
# NORA_API_KEY_HASH_SECRET_FILE=/run/secrets/NORA_API_KEY_HASH_SECRET
# In production the backend refuses to boot when JWT_SECRET is missing/placeholder
# or ENCRYPTION_KEY is missing/invalid. To deliberately run without encryption at
# rest (e.g. a throwaway demo), set the explicit override:
# NORA_ALLOW_PLAINTEXT_SECRETS=true

# ── Bootstrap Admin Account (required for empty PaaS; optional for self-hosted) ──
# On an empty user database, hosted PaaS requires both values. Self-hosted installs may blank both
# to use first-account claim. Setting either value makes this a startup gate: placeholders,
# malformed email, short passwords, and default-derived passwords are rejected.
DEFAULT_ADMIN_EMAIL=<REPLACE_WITH_BOOTSTRAP_ADMIN_EMAIL>
DEFAULT_ADMIN_PASSWORD=<REPLACE_WITH_STRONG_BOOTSTRAP_PASSWORD>   # min 12 chars; no insecure defaults

# ── Database (defaults work with Docker Compose) ─────────────
DB_HOST=postgres
DB_USER=nora
DB_PASSWORD=<REPLACE_WITH_YOUR_DB_PASSWORD>
# DB_PASSWORD_FILE=/run/secrets/DB_PASSWORD
DB_NAME=nora
DB_PORT=5432

# ── Redis (defaults work with Docker Compose) ────────────────
REDIS_HOST=redis
REDIS_PORT=6379
PORT=4000
BACKEND_API_PORT=4100                         # host-only backend API port; setup auto-picks another free port when 4100 is busy
# Numeric group owner of /var/run/docker.sock. setup auto-detects it so the
# non-root API/workers can use the Docker API without running as root.
DOCKER_GID=0
# Stable Compose namespace used by setup for scoped starts, upgrades, and
# destructive clean-reinstall cleanup.
COMPOSE_PROJECT_NAME=nora

# Managed/external service connection options. DATABASE_URL/REDIS_URL take
# precedence over the host/port fields above. Verification modes fail closed
# when their CA is missing.
DATABASE_URL=
DB_SSL_MODE=                                  # blank honors URL sslmode; or disable | require | verify-ca | verify-full
DB_SSL_CA=
DB_SSL_CA_FILE=
DB_SSL_CERT=
DB_SSL_CERT_FILE=
DB_SSL_KEY=
DB_SSL_KEY_FILE=
DB_POOL_MAX=20
DB_IDLE_TIMEOUT_MS=30000
DB_CONNECTION_TIMEOUT_MS=10000
DB_STATEMENT_TIMEOUT_MS=0
DB_MIGRATION_LOCK_TIMEOUT_MS=60000
DB_MIGRATION_STATEMENT_TIMEOUT_MS=600000
REDIS_URL=
REDIS_USERNAME=
REDIS_PASSWORD=
REDIS_DB=
REDIS_TLS=false
REDIS_TLS_CA=
REDIS_TLS_CA_FILE=
REDIS_TLS_CERT=
REDIS_TLS_CERT_FILE=
REDIS_TLS_KEY=
REDIS_TLS_KEY_FILE=
REDIS_TLS_INSECURE_SKIP_VERIFY=false
REDIS_CONNECT_TIMEOUT_MS=10000

# ── Deployment Queue / Provisioner throughput ──────────────
# Higher values reduce time spent in "queued" under bursty deploy load, but
# increase image-pull/bootstrap pressure on the host.
DEPLOYMENT_WORKER_CONCURRENCY=6

# ── Access / URL ─────────────────────────────────────────────
# Local default:
#   NGINX_CONFIG_FILE=nginx.conf
#   NGINX_HTTP_PORT=8080
#   NEXTAUTH_URL=http://localhost:8080
# Public domain:
#   NGINX_CONFIG_FILE=nginx.public.conf
#   NGINX_HTTP_PORT=80
#   NEXTAUTH_URL=https://app.example.com
#   CORS_ORIGINS=https://app.example.com
NGINX_CONFIG_FILE=nginx.conf
NGINX_HTTP_PORT=8080
# Force the Secure flag on the session cookie for always-on-TLS public deploys.
# Set to 1 for https public modes; leave empty for local http. Without it the
# Secure flag relies on X-Forwarded-Proto, which an upstream proxy/CDN may strip.
NORA_FORCE_SECURE_COOKIES=

# ── API Request Rate Limits (per IP) ───────────────────────
# App-wide abuse protection. Defaults below are the production values; raise
# them only where many requests legitimately share one source IP (e.g. CI/E2E
# behind a single localhost address). Unset means the default applies.
GLOBAL_RATE_LIMIT_MAX=1000                         # all requests per IP/window
GLOBAL_RATE_LIMIT_WINDOW_MS=900000                # 15 minutes
MUTATION_RATE_LIMIT_MAX=60                         # POST/PUT/PATCH/DELETE per IP/window
MUTATION_RATE_LIMIT_WINDOW_MS=60000               # 1 minute

# ── OAuth (disabled by default until provider verification is completed) ──
# Password and OAuth login share this per-IP abuse-protection budget.
AUTH_RATE_LIMIT_MAX=20
AUTH_RATE_LIMIT_WINDOW_MS=900000                  # 15 minutes
OAUTH_LOGIN_ENABLED=false
GOOGLE_CLIENT_ID=<REPLACE_WITH_GOOGLE_CLIENT_ID>
GOOGLE_CLIENT_SECRET=<REPLACE_WITH_GOOGLE_CLIENT_SECRET>
GITHUB_CLIENT_ID=<REPLACE_WITH_GITHUB_CLIENT_ID>
GITHUB_CLIENT_SECRET=<REPLACE_WITH_GITHUB_CLIENT_SECRET>
NEXTAUTH_URL=http://localhost:8080                # public base URL; kept for compatibility with existing deploy envs

# ── Public Signup Abuse Protection ──────────────────────────
# Signup stays public, but these app-level limits keep automated spikes bounded.
SIGNUP_RATE_LIMIT_BURST_MAX=5
SIGNUP_RATE_LIMIT_BURST_WINDOW_MS=600000          # 10 minutes
SIGNUP_RATE_LIMIT_DAILY_MAX=20
SIGNUP_RATE_LIMIT_DAILY_WINDOW_MS=86400000        # 24 hours
# Optional bot challenge. Leave provider as none for local/self-hosted default.
# Choose one provider at runtime: none | turnstile | recaptcha. The backend
# publishes only the selected provider and public site key to the signup page.
SIGNUP_BOT_PROTECTION_PROVIDER=none
SIGNUP_TURNSTILE_SITE_KEY=
SIGNUP_TURNSTILE_SECRET=                       # server-only; never exposed to the browser
SIGNUP_RECAPTCHA_SITE_KEY=
SIGNUP_RECAPTCHA_SECRET=                       # server-only; never exposed to the browser
# Deprecated NEXT_PUBLIC_SIGNUP_* aliases remain readable at runtime for
# existing installs, but new deployments should use the variables above.

# ── Marketing site analytics (optional, Plausible) ──────────
# Off by default — self-hosted deployments ship zero tracking. Set the domain
# to enable privacy-light analytics on the marketing site. Point ANALYTICS_SRC
# at a self-hosted Plausible instance instead of plausible.io if you run one.
NEXT_PUBLIC_ANALYTICS_DOMAIN=
# NEXT_PUBLIC_ANALYTICS_SRC=https://plausible.io/js/script.js

# ── Release Tracking / Admin Upgrade Banner (optional) ──────
# Set NORA_CURRENT_VERSION from your deploy pipeline. If NORA_GITHUB_REPO
# is set, Nora will fetch the latest published GitHub release automatically
# and compare it against the running version.
#
# Typical GitHub release notes URL:
#   https://github.com/<owner>/<repo>/releases/tag/v1.2.3
#
# Typical source-based self-hosted upgrade command:
#   ./setup.sh --update
NORA_CURRENT_VERSION=
NORA_CURRENT_COMMIT=
NORA_GITHUB_REPO=solomon2773/nora
NORA_GITHUB_TOKEN=                                # optional; helps with GitHub API rate limits
NORA_RELEASE_CACHE_TTL_MS=300000                 # 5 minutes
NORA_LATEST_VERSION=
NORA_LATEST_PUBLISHED_AT=
NORA_RELEASE_NOTES_URL=
NORA_LATEST_SEVERITY=warning                      # info | warning | critical
NORA_UPGRADE_REQUIRED=false
NORA_AUTO_UPGRADE_ENABLED=false                  # opt-in one-click upgrades from Admin Settings
NORA_HOST_REPO_DIR=                              # absolute Linux host path to this Nora checkout; setup fills this
NORA_UPGRADE_REPO=https://github.com/solomon2773/nora.git
NORA_UPGRADE_REF=master
NORA_UPGRADE_RUNNER_IMAGE=docker:29-cli
NORA_UPGRADE_STATE_VOLUME=nora_upgrade_state
NORA_ENV_FILE=.env                                 # deploy env file passed to docker compose
NORA_UPGRADE_COMPOSE_FILES=                       # colon-separated compose files; blank auto-detects current stack
NORA_UPGRADE_PUBLIC_HEALTH_URL=                   # optional public /api/health URL checked after rebuild
# Shared by setup post-start probes and one-click upgrade health checks.
# 221 attempts, 3s apart = 660s from the first to final attempt. Overrides
# must be positive integers with a first-to-final window no greater than 3900s.
NORA_UPGRADE_HEALTHCHECK_ATTEMPTS=221
NORA_UPGRADE_HEALTHCHECK_INTERVAL_SECONDS=3
NORA_INSTALL_METHOD=source                       # source | docker_image | custom
NORA_MANUAL_UPGRADE_COMMAND=./setup.sh --update
# Optional advanced override. Leave unset to use the built-in default steps.
NORA_MANUAL_UPGRADE_STEPS=

# ── System Banner / Testing Warning (optional) ───────────────
# Master switch for the admin-configured banner shown across /app and /admin.
# Banner copy and severity are managed from the admin settings page.
NORA_SYSTEM_BANNER_ENABLED=false

# ── Platform Mode ─────────────────────────────────────────────
# "selfhosted" (default) — operator sets max resource limits below;
#                           users pick vCPU/RAM/disk within those limits.
# "paas"                 — Stripe plans control agent-count and managed-backup
#                           entitlements; Admin deployment defaults supply one
#                           platform-wide vCPU/RAM/requested-disk profile.
PLATFORM_MODE=selfhosted

# ── Self-hosted resource limits (only when PLATFORM_MODE=selfhosted) ─
MAX_VCPU=16
MAX_RAM_MB=32768
MAX_DISK_GB=500
MAX_AGENTS=50

# ── Managed Backups ──────────────────────────────────────────
# NORA_BACKUP_STORAGE can be local, s3, r2, or ssh. Leave destination env
# values empty to use Admin Settings and the default local backup volume.
NORA_BACKUP_STORAGE=
NORA_BACKUP_DIR=
NORA_BACKUP_LIMIT_PER_AGENT=10
NORA_BACKUP_STORAGE_MB=51200
NORA_BACKUP_RETENTION_DAYS=30
BACKUP_WORKER_CONCURRENCY=2
NORA_BACKUP_JOB_TIMEOUT_MS=1800000
NORA_BACKUP_SCHEDULE_POLL_MS=60000
AGENT_PROVISION_LOCK_TIMEOUT_MS=30000

# S3-compatible storage. Use NORA_BACKUP_STORAGE=s3 for AWS S3.
NORA_BACKUP_S3_BUCKET=
NORA_BACKUP_S3_REGION=
NORA_BACKUP_S3_ENDPOINT=
NORA_BACKUP_S3_ACCESS_KEY_ID=
NORA_BACKUP_S3_SECRET_ACCESS_KEY=
NORA_BACKUP_S3_SESSION_TOKEN=

# Cloudflare R2. Set the account endpoint, for example:
# https://<account-id>.r2.cloudflarestorage.com
NORA_BACKUP_R2_BUCKET=
NORA_BACKUP_R2_REGION=
NORA_BACKUP_R2_ENDPOINT=
NORA_BACKUP_R2_ACCESS_KEY_ID=
NORA_BACKUP_R2_SECRET_ACCESS_KEY=
NORA_BACKUP_R2_SESSION_TOKEN=

# SSH/SFTP remote storage.
NORA_BACKUP_SSH_HOST=
NORA_BACKUP_SSH_PORT=
NORA_BACKUP_SSH_USERNAME=
NORA_BACKUP_SSH_REMOTE_PATH=
NORA_BACKUP_SSH_PRIVATE_KEY=
NORA_BACKUP_SSH_PASSWORD=

# ── Billing / Stripe (only when PLATFORM_MODE=paas) ─────────
BILLING_ENABLED=false
STRIPE_SECRET_KEY=<REPLACE_WITH_STRIPE_SECRET_KEY>              # sk_test_... or sk_live_...
STRIPE_WEBHOOK_SECRET=<REPLACE_WITH_STRIPE_WEBHOOK_SECRET>      # whsec_...
STRIPE_PRICE_PRO=<REPLACE_WITH_STRIPE_PRICE_ID>                 # price_...
STRIPE_PRICE_ENTERPRISE=<REPLACE_WITH_STRIPE_PRICE_ID>          # price_...

# ── Cost reporting (optional; used by /app/cost) ─────────────
COST_PER_1K_TOKENS=0.002
# Optional per-model token rates. Keys may be model ids or provider/model ids.
# Split rates are used when Nora records input/output token metadata; per_1k
# is used when only total tokens are available.
COST_MODEL_RATES_JSON={"openai/gpt-5.5":{"input_per_1k":0.002,"output_per_1k":0.008},"claude-sonnet-4-5":{"input_per_1k":0.003,"output_per_1k":0.015},"gemini-3.1-pro-preview":{"input_per_1k":0.00125,"output_per_1k":0.01}}

# ── OpenTelemetry GenAI export (optional; off by default) ────
# Export per-exchange chat spans + token/cost/resource metrics under the OTel
# gen_ai.* semantic conventions. Disabled unless NORA_OTEL_ENABLED=true, and
# fully fail-open (a collector outage never affects request handling).
NORA_OTEL_ENABLED=false
# OTLP/HTTP push target for metrics AND spans (e.g. an OpenTelemetry Collector,
# Grafana Alloy, or a vendor endpoint). OTEL_EXPORTER_OTLP_HEADERS is read
# natively by the exporters for auth.
OTEL_EXPORTER_OTLP_ENDPOINT=
# OTEL_EXPORTER_OTLP_HEADERS=authorization=Bearer%20<token>
OTEL_SERVICE_NAME=nora-control-plane
# Prometheus pull endpoint for metrics (0 or empty disables). Defaults to
# loopback — the scrape surface is unauthenticated and its labels (agent id /
# model / provider) are operational intelligence, so widen the bind only behind
# an internal network or firewall. Spans are OTLP-only (Prometheus carries
# metrics, not traces).
NORA_OTEL_PROMETHEUS_PORT=9464
NORA_OTEL_PROMETHEUS_HOST=127.0.0.1

# ── Scheduled agent runs (recurring cron triggers) ──────────
# Minimum seconds between fires for any schedule — a guardrail so a too-frequent
# cron can't thrash an agent (and its budget). Default 60s.
NORA_SCHEDULE_MIN_INTERVAL_SECONDS=60
# Worker concurrency for executing due scheduled runs (default 5).
SCHEDULE_RUN_WORKER_CONCURRENCY=5

# ── Runtime families, deploy targets, and sandbox profiles ───
# High-level runtime contracts. Supported ids: openclaw, hermes.
ENABLED_RUNTIME_FAMILIES=openclaw
#
# Execution target ids. Supported ids: docker, proxmox.
# Local no-cluster fallback:
ENABLED_BACKENDS=docker
# Docker-published agent gateway/runtime ports bind to loopback by default.
# This governs ALL local Docker runtime families (OpenClaw gateway/runtime and
# Hermes runtime API 8642 + dashboard 9119). To let an external client — e.g.
# Hermes Desktop over Tailscale — reach an agent, set this to a routable host
# interface (your Tailscale IP, e.g. 100.71.115.105). Never use 0.0.0.0 without
# a firewall or authenticated reverse proxy in front.
DOCKER_AGENT_BIND_IP=127.0.0.1
# Kubernetes clusters are registered in Admin -> Kubernetes; enabled registry
# rows appear as execution targets like k8s:aks-eastus2.
#
# Sandbox profile ids. Supported ids: standard, nemoclaw.
ENABLED_SANDBOX_PROFILES=standard

# Kubernetes Compose overlay mounts kubeconfigs only. Provider, namespaces,
# exposure, annotations, source ranges, and load-balancer class are configured
# in Admin -> Kubernetes.
# Kubeconfigs should be readable by Docker Compose but must not be committed;
# .secrets/ is gitignored for this purpose.
#
# Mounted-path mode:
#   mkdir -p ./.secrets/kubeconfigs
#   cp ./.secrets/aks-kubeconfig ./.secrets/kubeconfigs/aks-eastus2
#   cp ./.secrets/aks-westus2-kubeconfig ./.secrets/kubeconfigs/aks-westus2
# Then use /kubeconfigs/<file> as each Admin Kubeconfig path. For one
# cluster, put one file in this directory and register one Admin row.
NORA_KUBECONFIGS_DIR=./.secrets/kubeconfigs

# Docker and remote-Docker agents reserve published gateway ports from this
# range. Values must stay within 19000-19999, which is the default gateway-proxy
# security envelope. Local Docker also skips ports already published by running
# containers and retries bounded bind conflicts from non-Docker listeners.
DOCKER_AGENT_PORT_RANGE_MIN=19000
DOCKER_AGENT_PORT_RANGE_MAX=19999

# Gateway proxy SSRF guard overrides. Defaults allow OpenClaw's internal
# gateway port, Docker-published 19000-19999, and Kubernetes NodePort 30000-32767.
NORA_GATEWAY_PROXY_ALLOWED_PORTS=
NORA_GATEWAY_PROXY_ALLOWED_HOSTS=

# Active Remote Docker gateway, log, and terminal streams periodically re-check
# the agent owner's current host grant. Values below 250 ms are clamped to 250.
REMOTE_HOST_AUTH_RECHECK_MS=1000

# Serialize create, edit, Test, SSH-pin reset, delete, share, and unshare for
# one Remote Docker host id.
# A timed-out competing mutation returns 409 instead of racing trust state.
REMOTE_HOST_MUTATION_LOCK_TIMEOUT_MS=15000

# ── Proxmox LXC (experimental; opt in with ENABLED_BACKENDS) ────────────
# OpenClaw works with a stock Ubuntu LXC template. Hermes requires a prepared
# template. Run the real-host smoke before production use. NemoClaw on Proxmox
# remains blocked because LXC does not implement its OpenShell sandbox contract.
PROXMOX_API_URL=<REPLACE_WITH_PROXMOX_URL>          # https://proxmox.local:8006/api2/json
PROXMOX_TOKEN_ID=<REPLACE_WITH_TOKEN_ID>            # user@pam!tokenname
PROXMOX_TOKEN_SECRET=<REPLACE_WITH_TOKEN_SECRET>
PROXMOX_VERIFY_TLS=true
PROXMOX_CA_CERT=
PROXMOX_CA_CERT_PATH=
# Test-only escape hatch for isolated HTTP fixtures. Never use in production.
PROXMOX_ALLOW_INSECURE_HTTP=false
PROXMOX_NODE=pve
PROXMOX_TEMPLATE=local:vztmpl/ubuntu-22.04-standard_22.04-1_amd64.tar.zst
PROXMOX_HERMES_TEMPLATE=
PROXMOX_ROOTFS_STORAGE=local-lvm
PROXMOX_BRIDGE=vmbr0
PROXMOX_SSH_HOST=<REPLACE_WITH_PROXMOX_SSH_HOST>
PROXMOX_SSH_USER=root
PROXMOX_SSH_PORT=22
PROXMOX_SSH_PRIVATE_KEY=
PROXMOX_SSH_PRIVATE_KEY_PATH=
PROXMOX_SSH_PRIVATE_KEY_PASSPHRASE=
PROXMOX_SSH_PASSWORD=
PROXMOX_SSH_HOST_FINGERPRINT=                    # ssh-keyscan + ssh-keygen -lf output, SHA256:...
# Test-only escape hatch. Production must pin PROXMOX_SSH_HOST_FINGERPRINT.
PROXMOX_SSH_INSECURE_ACCEPT_HOST_KEY=false
# One command name or absolute executable path; arguments are rejected.
PROXMOX_PCT_COMMAND=pct
# Leave empty for root SSH. Non-root defaults to sudo -n; an override must be
# exactly sudo -n or an absolute sudo path followed by -n.
PROXMOX_SUDO=
# Root SSH is the supported baseline. Non-root SSH users must install a
# separately audited, narrowly scoped stopped-LXC staging helper on the
# Proxmox host and configure its one absolute executable path here.
PROXMOX_OFFLINE_STAGE_COMMAND=
PROXMOX_NODE_MAJOR=24
PROXMOX_OPENCLAW_PACKAGE=openclaw@2026.6.11
PROXMOX_HERMES_BIN=/opt/hermes/.venv/bin/hermes
# Hermes dashboard stays loopback-only unless this explicit insecure opt-in is set.
PROXMOX_HERMES_ENABLE_INSECURE_DASHBOARD=false

# ── NemoClaw / NVIDIA (when ENABLED_SANDBOX_PROFILES includes nemoclaw) ──
NVIDIA_API_KEY=                                     # from build.nvidia.com
NEMOCLAW_DEFAULT_MODEL=nvidia/nemotron-3-super-120b-a12b
# Default: Nora-built image layering tsx + sandbox-writable runtime dirs on
# top of the upstream OpenShell base (agent-runtime/Dockerfile.nemoclaw-agent).
# Defaults to the Nora-published GHCR image. For offline hosts or private
# clusters, build/preload nora-nemoclaw-agent:local and override this value.
NEMOCLAW_SANDBOX_IMAGE=ghcr.io/solomon2773/nora-nemoclaw-agent:latest

# Nora's validated OpenClaw package. Override only for a deliberate compatibility test.
OPENCLAW_DOCKER_PACKAGE=openclaw@2026.6.11

# ── Security ─────────────────────────────────────────────────
CORS_ORIGINS=http://localhost:8080                # comma-separated; use your public origin when exposed on a domain

# ── LLM Key Storage: database (default) | env ───────────────
KEY_STORAGE=database
# If KEY_STORAGE=env, set provider credentials directly in this file instead
# of using Settings. Microsoft Foundry also needs the per-resource URL and
# deployment name because Azure deployments are operator-defined.
# MICROSOFT_FOUNDRY_API_KEY=
# MICROSOFT_FOUNDRY_BASE_URL=https://<resource>.services.ai.azure.com/openai/v1/
# MICROSOFT_FOUNDRY_DEPLOYMENT=gpt-5.5-1
# MICROSOFT_FOUNDRY_API_VERSION=

# ── Backups & TLS (optional) ─────────────────────────────────
TLS_CERT_PATH=
TLS_KEY_PATH=
AWS_S3_BUCKET=<REPLACE_WITH_S3_BUCKET_NAME>
AWS_ACCESS_KEY_ID=<REPLACE_WITH_AWS_ACCESS_KEY>
AWS_SECRET_ACCESS_KEY=<REPLACE_WITH_AWS_SECRET_KEY>

# When true, the provisioner reconciler uninstalls ClawHub skills present in a runtime
# container but not tracked in the agents table. OFF by default — leaving it unset only
# surfaces such drift as "orphaned" in the UI; operators remove orphans explicitly.
CLAWHUB_PRUNE_ORPHANED_SKILLS=false