# Use a server-side Syntarus project key. Never expose it to clients.
SYNTARUS_API_KEY=sk_mem_replace_me
# For a secret manager/Docker secret, use SYNTARUS_API_KEY_FILE instead.
SYNTARUS_BASE_URL=https://ai.syntarus.com/v1
SMARA_DATABASE_URL=postgresql://smara:change-me@postgres:5432/smara
SMARA_DATABASE_PATH=./data/smara.db
SMARA_POSTGRES_PASSWORD=change-me
SMARA_COMPOSE_PROJECT=smara
SMARA_BIND_ADDRESS=127.0.0.1
SMARA_HOST_PORT=8090
# Development only. Keep this false in any shared or deployed environment;
# production requests must arrive through a verified native Smara session,
# device token, or server-to-server assertion.
SMARA_DEV_MODE=false
# Optional server-to-server gateway assertion secret for non-browser clients.
# It is never shipped to the frontend and is not a user credential.
SMARA_GATEWAY_SIGNING_SECRET=replace-with-a-long-random-secret
# Optional compatibility secret for an internal bearer assertion. Native Web
# auth uses SMARA_SESSION_SECRET and does not require this value.
SMARA_CONTROL_BRIDGE_SECRET=replace-with-a-different-long-random-secret
# Native Smara auth. Point this at the existing accounts database during the
# cutover so account ids (and therefore memory namespaces) are preserved.
SMARA_ACCOUNTS_DATABASE_URL=postgresql://memoryos:change-me@memoryos-postgres:5432/memoryos_accounts
SMARA_SESSION_SECRET=replace-with-a-long-random-session-secret
SMARA_SESSION_TTL_DAYS=30
SMARA_AUTH_COOKIE_NAME=smara_session
# Optional private operator console. Leave unset to keep /admin unavailable.
# The browser receives only a short-lived httpOnly session cookie.
SMARA_OPERATOR_SECRET=replace-with-a-long-random-operator-secret
SMARA_OPERATOR_COOKIE_NAME=smara_operator
SMARA_OPERATOR_SESSION_HOURS=12
SMARA_INTERNAL_TOKEN=replace-with-a-long-random-worker-token
# Browser origins allowed to call Smara.
SMARA_ALLOWED_ORIGINS=https://ai.syntarus.com
# Hosted control-plane policy: personal integrations and browser sessions stay
# on the paired desktop. Set true only for a separately reviewed migration.
SMARA_HOSTED_USER_INTEGRATIONS_ENABLED=false
# Legacy hosted integration vault. Leave these empty in the default hosted
# control-plane deployment: user credentials and browser sessions stay on the
# paired desktop. Set the explicit opt-in flag only for a reviewed migration.
SMARA_INTEGRATION_MASTER_KEY=
# Rotation: prepend the new key, keep the old one until all credentials have
# been read and re-encrypted by the integration worker, then remove it.
SMARA_INTEGRATION_MASTER_KEYS=
SMARA_PUBLIC_BASE_URL=https://ai.syntarus.com
# Required only when enabling the matching OAuth connection buttons.
SMARA_GOOGLE_CLIENT_ID=
SMARA_GOOGLE_CLIENT_SECRET=
SMARA_TELEGRAM_BOT_TOKEN=
SMARA_TELEGRAM_BOT_URL=
SMARA_TELEGRAM_CHAT_URL=http://api:8080
SMARA_GITHUB_CLIENT_ID=
SMARA_GITHUB_CLIENT_SECRET=
# Enable Web Push for the installed phone PWA. Generate/store these only in
# deployment secrets; the public key is intentionally safe to return to phones.
SMARA_VAPID_PUBLIC_KEY=
SMARA_VAPID_PRIVATE_KEY=
SMARA_VAPID_SUBJECT=mailto:admin@example.com
SMARA_RATE_LIMIT_PER_MINUTE=120
# Required outside development: all API replicas use this Redis bucket.
SMARA_REDIS_URL=redis://redis:6379/0
# Optional error reporting. No events leave this process when it is empty.
SMARA_SENTRY_DSN=
# Direct-chat provider for the first extracted Memento runtime slice. Any
# OpenAI-compatible endpoint is supported. Leave blank to keep /v1/chat safely
# unavailable rather than silently using an unconfigured provider.
SMARA_LLM_BASE_URL=https://api.example.com/v1
SMARA_LLM_API_KEY=
SMARA_LLM_MODEL=

# Fair GAIA evaluation (opt-in; readiness does not contact the network)
SMARA_GAIA_DATASET=
SMARA_BENCHMARK_MODEL_ENDPOINT=
SMARA_BENCHMARK_MODEL=
SMARA_BENCHMARK_API_KEY=
SMARA_BENCHMARK_AUTH_HEADER=authorization
SMARA_LLM_PROVIDER=configured model provider
# Optional allowlisted profiles. Keys stay server-side; clients send only a
# profile name. Sarvam uses its native api-subscription-key header. GLM-5.2
# and Gemma 4 are beta/entitlement-gated on Sarvam; keep Grok as the default
# until the operator has verified access.
SMARA_XAI_KEY=
SMARA_SARVAM_KEY=
# {"grok":{"base_url":"https://api.x.ai/v1","model":"grok-4.6","api_key_env":"SMARA_XAI_KEY","auth_header":"Authorization","capability":"chat"},"sarvam":{"base_url":"https://api.sarvam.ai/v1","model":"sarvam-105b","api_key_env":"SMARA_SARVAM_KEY","auth_header":"api-subscription-key","capability":"chat"},"sarvam-reasoning":{"base_url":"https://api.sarvam.ai/v2","model":"glm5.2","api_key_env":"SMARA_SARVAM_KEY","auth_header":"api-subscription-key","capability":"reasoning"},"sarvam-vision":{"base_url":"https://api.sarvam.ai/v2","model":"gemma4","api_key_env":"SMARA_SARVAM_KEY","auth_header":"api-subscription-key","capability":"vision"}}
SMARA_LLM_PROFILES=
SMARA_LLM_DEFAULT_PROFILE=grok
# Declarative plugin/MCP descriptors only; arbitrary Python is never imported.
SMARA_PLUGIN_MANIFESTS=
# Sandbox is disabled unless a separately isolated service is deployed.
SMARA_SANDBOX_ENABLED=false
SMARA_SANDBOX_URL=
SMARA_SANDBOX_TOKEN=
# Optional and opt-in. When enabled, research reports are synthesized only
# from verified evidence with citation validation. Missing/failed providers
# fall back to the deterministic cited report.
SMARA_RESEARCH_SYNTHESIS_ENABLED=false
# Separate signing secret for CLI bearer tokens. Never reuse the browser bridge
# or gateway secret. Rotate by invalidating/reissuing CLI tokens at rollout.
SMARA_CLI_TOKEN_SECRET=
SMARA_CLI_TOKEN_TTL_DAYS=30
# Provider-neutral web research. Keep the provider key server-side; it is
# never returned to the Web, CLI, desktop, or task artifacts.
# brave, serper, tavily, or exa. The URL is optional and defaults per provider.
SMARA_SEARCH_PROVIDER=brave
SMARA_SEARCH_API_KEY=
SMARA_SEARCH_URL=https://api.search.brave.com/res/v1/web/search
# Optional independent fallback. Deep research retries all search angles here
# when the primary provider is unavailable or lacks source diversity.
SMARA_SEARCH_FALLBACK_PROVIDER=
SMARA_SEARCH_FALLBACK_API_KEY=
SMARA_SEARCH_FALLBACK_URL=
SMARA_SEARCH_TIMEOUT_SECONDS=12
# Deep research deliberately limits parallel provider calls to avoid losing
# evidence to provider burst limits.  Use 1 for a strict low-rate plan.
SMARA_SEARCH_MAX_CONCURRENCY=2
# Optional source policy. Empty allow-list means public domains are allowed;
# blocked domains always win over allowed domains.
SMARA_RESEARCH_ALLOWED_DOMAINS=
SMARA_RESEARCH_BLOCKED_DOMAINS=
# Capture providers are optional and can reuse SMARA_SARVAM_KEY through the
# server-side fallback in config.py. Voice keeps the existing
# OpenAI-compatible transcription contract; configure it only when that
# endpoint is available. Photo analysis uses Gemma 4 and PDF/image OCR uses
# Sarvam Document AI.
SMARA_CAPTURE_TRANSCRIPTION_BASE_URL=
SMARA_CAPTURE_TRANSCRIPTION_API_KEY=
SMARA_CAPTURE_TRANSCRIPTION_MODEL=
SMARA_CAPTURE_TRANSCRIPTION_AUTH_HEADER=Authorization
SMARA_CAPTURE_VISION_BASE_URL=https://api.sarvam.ai/v2
SMARA_CAPTURE_VISION_API_KEY=
SMARA_CAPTURE_VISION_MODEL=gemma4
SMARA_CAPTURE_VISION_AUTH_HEADER=api-subscription-key
SMARA_CAPTURE_OCR_BASE_URL=https://api.sarvam.ai/doc-ai/v1
SMARA_CAPTURE_OCR_API_KEY=
SMARA_CAPTURE_OCR_MODEL=sarvam-vision-v1
SMARA_CAPTURE_OCR_LANGUAGE=en-IN
SMARA_CAPTURE_OCR_AUTH_HEADER=api-subscription-key
# Performance rollout switches. Keep enabled after the beta acceptance suite
# is green; any single optimization can be disabled without a code deploy.
SMARA_FAST_ROUTING_ENABLED=true
SMARA_POOLED_RESOURCES_ENABLED=true
SMARA_WORK_SIGNALS_ENABLED=true
SMARA_DESKTOP_LONG_POLL_ENABLED=true
SMARA_SHADOW_ROUTING_ENABLED=false
SMARA_WORKER_CONCURRENCY=4