Skip to content
Configuration reference

Configuration reference

This reference was built from the code. It lists the environment variables an operator sets on the FACE backend, the platform libraries it links, and the frontend. It leaves out variables read only by build tooling, benchmarks and tests, variables read only by the separate voice calling-agent image, and internal tuning knobs whose defaults are meant to stay as shipped.

In a cluster, set control-plane values through the ControlPlane resource’s env list, never on the live Deployment, and runner values in the face-runner manifest (see Deployment model). Locally, export them into the backend’s environment before you start it.

Secret in the Source column means the value is resolved through the platform secret backend (SECRET_BACKEND), so it can come from a mounted file instead of the environment. Everything else is read from the environment.

Connector and integration credentials (warehouses, SaaS APIs, maps keys) do not belong in environment variables. They are entered per connection in Settings › Integrations and stored sealed in FACE’s credential store. The few integration variables below are the exceptions the code still reads.

Required at boot

VariableSourcePurpose
AUTH_JWT_SECRETsecretSigns the pre-TOTP MFA challenge. The process refuses to start without it.
MESH_CA_CERTsecretThe shared cluster CA certificate. Each pod issues its own rotating mTLS leaves from it.
MESH_CA_KEYsecretThe shared cluster CA private key.
TIDE_ENVELOPE_KEKsecretA base64-encoded 32-byte key-encryption key for everything sealed at rest.
INFERENCE_REMOTE_URLenvThe base URL of the shared inference plane. No other name is read for it.

Prerequisites and ordering has the exact refusal messages.

Process, role and platform

VariableDefaultPurpose
SERVICE_ROLEcontrol planerunner makes this process a fetch runner. Any other value makes it a control plane. The operator sets server or runner.
PORTnoneThe default for serve --port. The image entrypoint uses 7000 when PORT is unset. The operator sets 7100 (control plane) or 7102 (runner).
APP_VERSIONemptyThe version stamped on structured logs.
SECRET_BACKENDenvenv or file. With file, secrets are read from SECRETS_DIR with the environment as fallback. Any other value makes every secret lookup fail.
SECRETS_DIR/var/run/secretsThe directory for the file backend. A file is named after the variable, or in lower case with dashes.
RUNINK_VARIANTstandardThe declared variant, reported to the FACE app. It changes no data and no behaviour on any variant: FACE creates no connections and serves no rows of its own.
FACE_LOCAL_INSTANCEfalsetrue offers a single local-development instance in the instance picker, on a process with no in-cluster ServiceAccount token only. In a cluster it changes nothing.
RUNINK_APPfaceThe app name that filters ClientInstances in the instance picker.
FACE_CLIENT_INSTANCEnoneThe name of this deployment’s own ClientInstance, used by the access page. There is no fallback.
FACE_CLIENT_INSTANCE_NAMESPACEPOD_NAMESPACEThe namespace of that ClientInstance.
POD_NAMESPACEnoneThe pod’s namespace, from the downward API. Also used when publishing connectors to TIDE.
POD_NAMEnoneThe instance name on lineage reports sent to TIDE.
HOSTNAMEset by KubernetesThe runner id fallback when RUNNER_ID is unset.
KUBERNETES_SERVICE_HOST, KUBERNETES_SERVICE_PORTkubernetes.default.svc, 443The API server FACE lists ClientInstances from.
FACE_CONNECTORS_APPfaceThe app key of the connectors ConfigMap FACE publishes to TIDE, <key>-connectors. TIDE sets it on each instance to the exact key it created and granted. Unset (the control plane), FACE writes face-connectors. A value that is not a DNS-1123 label (at most 63 characters of a-z, 0-9 and -, starting and ending with a letter or digit) is logged and nothing is published.
SUBSCRIPTION_TIER, SUBSCRIPTION_STATUS, SUBSCRIPTION_SEATSnoneThe subscription the operator mirrors from the ClientInstance. A subscription is reported only when the tier is set and FACE_TENANT matches the tenant asked about.
RUNINK_ACCOUNTS_FILEnoneThe account file the platform operator provides, mapping each user to the purchaser whose billing account pays for their model time. Unset, every model call is recorded as unattributed and logged, never billed to another account. See Plans and billing.
TIDE_ROUTER_BILLING_KEY_FILE, TIDE_ROUTER_BILLING_KIDnoneThe key file (read by path) and key id FACE signs each model call’s billing claim with. Unset, model calls carry no billing claim. It never blocks or fails a call.
TIDE_ROUTER_IDENTITY_KEYnoneThe model router’s key (base64, at least 32 bytes) FACE signs an identity assertion with for a fetch a person started, so it runs on the router’s interactive lane. It is never logged or sent to a client. Unset, every model call stays on its billing lane.
TIDE_PLATFORM_ADMINnoneThe installation admin’s e-mail. That account sees every instance. When it is unset, nobody gets that grant.
LICENSEnoneA signed licence blob, as base64 or raw JSON.
LICENSE_FILEnoneA path to read the licence from when LICENSE is unset.
LICENSE_PUBLIC_KEYnoneThe base64 Ed25519 public key licences are verified against.
FACE_MODEL_CAP_FETCH_PER_HOUR, FACE_MODEL_CAP_ASK_PER_HOUR, FACE_MODEL_CAP_LOOP_PER_HOUR, FACE_MODEL_CAP_SIMULATE_PER_HOUR, FACE_MODEL_CAP_VISION_PER_HOURset by FACEHow many model-backed runs one user may start per hour, per class: fetches (RunFetch, RunFetchStream), QuickAsk, agent loops (ExecuteLoop), simulations (SimulateScenario, InteractiveSimulate) and vision (ExtractText, IngestSurveillanceFeed). A call over the cap gets RESOURCE_EXHAUSTED with the time until the next slot.
TIDE_HEALTH_URL, TIDE_HEALTH_TOKENnoneWhere FACE reports observed lineage to the TIDE console, and the bearer token it uses.
CORS_ALLOWED_ORIGINSunsetA comma-separated allowlist of browser origins for gRPC-web. Set it to the origin(s) the FACE app is served from.
FACE_RATE_RPS, FACE_RATE_BURSTset by FACEConfigure the per-user request rate limit (sustained rate and burst). Callers over the limit get RESOURCE_EXHAUSTED. A value that does not parse keeps the default.
ENV, GCP_REGIONemptyLabels printed into the system state that compliance and domain analyses read.

Authentication

VariableSourceDefaultPurpose
ADMIN_EMAILenvnoneThe admin account created at boot.
ADMIN_PASSWORDsecretnoneIts password. If either value is missing, password admin login is disabled.
AUTH_GOOGLE_ONLYenvunsettrue makes Google sign-in the sign-in method; password sign-in shows Password sign-in is disabled — use Google sign-in.
FACE_REQUIRE_TOTPenvunset (off)Requires a TOTP second factor for password sign-in. all covers every account; a comma-separated list of roles (admin, writer, reader) covers only those roles. A covered account with no TOTP enrolled is not signed in by its password: the FACE app shows a setup key, and the first code from the authenticator app enrols the key and completes sign-in. Google sign-in relies on the identity provider’s own second factor. An unrecognised value stops the process.
AUTH_SEED_USERSenvnoneComma-separated email=role pairs to seed. The operator projects them from ClientInstance users.
AUTH_USERS_PATHenvthe data directoryA legacy users file. It is imported once into the accounts table and then sealed as <path>.migrated.
OIDC_ISSUERenvnoneThe OIDC issuer. Single sign-on is on only when both this and OIDC_CLIENT_ID are set.
OIDC_CLIENT_IDenvnoneFACE’s OAuth client id, and the default token audience. It must equal the FACE app’s google_client_id (the face-frontend-config ConfigMap).
OIDC_AUDIENCEenvOIDC_CLIENT_IDAn explicit audience override.
OIDC_JWKS_URLenvdiscoveredAn explicit JWKS URL instead of discovery.
AUTH_ALLOWED_EMAILSenvnoneThe single sign-on allowlist. When SSO is configured and this is empty, every SSO login is refused (SSO is not available on this instance). An unlisted account gets This account is not authorized for this instance. The operator projects it from the face-oauth Secret. Changes take effect when the pod restarts.
FACE_RECAPTCHA_SITE_KEYenvnoneThe reCAPTCHA v3 site (public) key for login. The FACE app reads it from GET /auth/config. Configure it with FACE_RECAPTCHA_SECRET_FILE to captcha-check logins.
FACE_RECAPTCHA_SECRET_FILEsecret pathnonePath to a file holding the reCAPTCHA secret key (a mounted Secret). It is read at each login, so a rotated Secret needs no restart.
FACE_RECAPTCHA_HOSTNAMESenvnoneComma-separated FACE host(s) a token must have been minted on. Required once a key is present. With a key present but the configuration incomplete, every login is refused until it is complete.
FACE_RECAPTCHA_MIN_SCOREenv0.5The reCAPTCHA v3 score a login must reach.

The login captcha is off while no site key and no secret file are present, which is the default: the operator projects these three from the optional face-recaptcha Secret, and reCAPTCHA is not enforced on FACE (owner, 2026-11-02).

Inference and the model plane

VariableDefaultPurpose
INFERENCE_REMOTE_URLnone (required)The base URL for chat completions.
INFERENCE_API_KEYemptySecret. Sent as Authorization: Bearer when it is set. FACE never generates one. The placeholder application-generated-key counts as empty.
VISION_REMOTE_URLINFERENCE_REMOTE_URLThe base URL for vision calls.
VOICE_REMOTE_URLINFERENCE_REMOTE_URLThe base URL for voice calls.
INFERENCE_CONTEXT_SIZE16384The prompt window, in tokens, that the agent loop budgets against.
INFERENCE_DECODE_TOK_Snone in FACE; the operator sets 1.8The fallback decode rate of the text tier, which sizes model-call budgets. Since 2026-10-26 (owner) FACE first uses the rate the model router publishes, then the rate measured from the plane’s own answers, and only then this value. A fetch with none of the three sends no model call and says why.
INFERENCE_VISION_DECODE_TOK_Snone in FACE; the operator sets 1.0The same fallback, for the vision tier.
FETCH_DECODE_TOK_S, REACT_DECODE_TOK_S, POSTURE_DECODE_TOK_Sthe text tier’s ratePin the decode rate for the fetch, the agent loop’s turn budgets and posture analysis. An operator’s pin wins over the measured rate.
FETCH_DEADLINE_SECONDS300The end-to-end fetch budget, in whole seconds. The hard ceiling is 900 seconds; a larger value is clamped to it, with a log line.
FETCH_PHASE_DATA_PREP_SECONDS, FETCH_PHASE_POSTURE_SECONDS, FETCH_PHASE_RULES_SECONDS, FETCH_PHASE_TRENDS_SECONDS, FETCH_PHASE_HYPOTHESES_SECONDS, FETCH_PHASE_SYNTHESIS_SECONDS30, 90, 90, 90, 30, 45 at the 300-second deadlinePer-phase caps of a fetch (data profiling, business domains, business rules, market trends, what-if scenarios, recommendations). Unset, each cap scales with the deadline. The older names POSTURE_LLM_TIMEOUT_SECONDS and METASEARCH_BUDGET_SECONDS are still read for the domains and trends phases when the new name is unset.
COMPLIANCE_LLM_BUDGET_SECONDS15The model time budget for compliance analysis.
FETCH_TREND_MAX_QUERIES2How many scenario market topics one fetch may search (1–6). Topics shared by the most scenarios go first; confirmed public names, when set, replace them.
FACE_ADMIT_MAX0 (off)The maximum number of concurrent inference calls. 0 dispatches directly with no scheduling. The operator sets 1.
FACE_ADMIT_QUEUE0 (unbounded)The maximum number of waiters before backpressure. The operator sets 32.
FACE_ADMIT_MODErejectreject or degrade when the queue is full.
FACE_ADMIT_WEIGHTSnoneTier weights, for example interactive:4,batch:1.
FACE_ADMIT_COALESCEofftrue lets identical concurrent prompts share one generation.
FACE_GATE_MAX_INFLIGHToffThe concurrent-submission cap of the inference gate.
FACE_GATE_FIRST_TOKEN_SECONDSoffThe longest acceptable queue wait. The gate is on only when this or FACE_GATE_MAX_INFLIGHT is positive.
FACE_GATE_DECODE_RATEnoneThe measured tokens per second the gate assumes.
FACE_GATE_REQUEST_SECONDSoffThe per-request time budget for inference calls.

These variables are read into the process configuration for compatibility. FACE starts no model server of its own, so they neither choose nor tune what the plane serves. Set the model on the plane instead. They are: INFERENCE_PORT, INFERENCE_HOST, INFERENCE_MODEL, INFERENCE_TIMEOUT, INFERENCE_PARALLEL, INFERENCE_CACHE_RAM, INFERENCE_CPU_STRICT, INFERENCE_THREADS, INFERENCE_BATCH, INFERENCE_GPU_LAYERS, INTERNAL_INFERENCE_IP, INTERNAL_BRIDGE_IP, INTERNAL_SUBNET, MODEL_DIR, MODEL_VOXTRAL, MODEL_WHISPER and MODEL_VISION. (MODEL_VISION is also printed in the boot line that names the vision tier.)

MODEL_INFERENCE_TTS and MODEL_WAV_TOKENIZER are no longer read. Setting them does nothing.

MODEL_VISION is a routing key, not a choice of weights. The model router selects a backend tier from this name and then rewrites the field to default before dispatch, so the value never reaches an engine — it only decides which engine. Use a logical tier token (vision); a weights filename classifies as the general TEXT tier, which answers an image request at HTTP 200 without seeing the image.

Embeddings

FACE does not embed. EMBEDDING_URL, MODEL_EMBEDDING_DIM, EMBEDDING_FORMAT and MODEL_EMBEDDING are not read, and setting them changes nothing. Retrieval is grep over plain text: see Grounding.

Storage and persistence

VariableSourceDefaultPurpose
OBJECTSTORE_ENDPOINTenvunsetThe object store (host:port or https://host:port). When it is unset, the appfs root uses an encrypted local directory. An endpoint without TLS is accepted only for a loopback or cluster-internal name.
OBJECTSTORE_ACCESS_KEY, OBJECTSTORE_SECRET_KEYsecretnoneObject-store credentials. They are required when an endpoint is set.
OBJECTSTORE_VACUUM_ACCESS_KEY, OBJECTSTORE_VACUUM_SECRET_KEYsecretnoneAn optional second credential used for every delete. Set both or neither.
OBJECTSTORE_BUCKETenvthe store library’s fallback; the deployment sets runink-face-db-nane1The object-store bucket for records, knowledge and evidence. Always set it explicitly. The appfs root and the documents mount use their own buckets (FACE_APPFS_BUCKET, FACE_DOCUMENTS_BUCKET).
FACE_APPFS_BUCKETenvfaceThe object-store bucket of the appfs root. TIDE sets one per instance. It must be 3 to 63 characters of a-z, 0-9 and -, starting and ending with a letter or digit; any other value stops FACE from starting.
FACE_DOCUMENTS_BUCKETenvface-documentsThe object-store bucket of the documents mount, with the same rule.
TIDE_ENVELOPE_KEK_RETIREDsecretnoneRetired KEKs (comma- or space-separated, base64), accepted for decryption only.
SQLITE_PATH<data dir>/face.dbA legacy face.db from an earlier build, imported once if it exists and never created. It also anchors the defaults below that say “beside face.db”. The operator does not set it on the control plane; the face-runner manifest sets /data/face-runner.db, which a runner never opens as a database.
FACE_APPFS_DIRappfs/ beside face.dbThe encrypted local appfs root, used only when there is no object store.
FACE_UPLOAD_DIRuploads/ beside face.dbThe instance’s private upload directory, created mode 0700. An OCR LOCAL source and a fetch attachment named by file name are read only from here, and uploaded attachments are written here. It may not be /tmp or another shared system directory, and there is no temp-directory fallback.
FACE_CCTV_CAPTURE_DIRSnoneExtra absolute directories (colon-separated) in which a camera connection may name a local video file. The upload directory is always included. Any other local path is refused.
FACE_TENANTdefault-tenantThe instance’s tenant, and the only tenant variable FACE reads. It names the tenant sub-root tenants/<tenant>/, the tenant sessions carry, and the tenant the access page and subscription checks compare against. It does not name the connectors ConfigMap (see FACE_CONNECTORS_APP). Set it before the first boot (see Persistence).
RUNNER_IDHOSTNAMEOn a runner, names its private tables. The face-runner manifest sets the stable value face-runner.
FACE_CONNECTIONS_DIRnoneThe legacy connection directory. It is imported once into the appfs root.
FACE_DCI_CORPUS_DIRnoneAn operator directory grounding also greps. FACE reads it and never writes to it. FACE’s own knowledge corpus is searched whether or not this is set.
FETCH_LEDGER_DIR./data/fetch-ledgerThe fetch cascade ledger.
RALPH_RUN_DIR./data/ralph-runsThe agent run journal, used to resume runs.
FACE_VOICE_SOURCE_DIR$RALPH_RUN_DIR/voice-sourcesWhere recorded voice sources are written.
RALPH_READ_ROOT.The root agents may read files under. Paths that escape it are rejected.
FACE_DEPLOY_SECRETS_DIRnoneA directory of runner deploy secrets, read by path.
EVIDENCE_FRAME_RETENTION0 (none kept)How long camera evidence frames are kept, as a Go duration. It is capped at 720h.

FACE persists nothing in SQLite (see Persistence).

Consensus, mesh and resilience

VariableDefaultPurpose
RAFT_NODE_IDunsetThis node’s consensus id. When it is unset, the process does not join consensus (runners). When it is set, a failure to start the orchestrator stops the process.
RAFT_PEERSnoneThe peers, as id@host:port,….
RAFT_PIPE_DIRnoneThe local proposal-pipe directory.
RAFT_DATA_DIRunset (in memory)The durable consensus state. When it is unset, FACE warns that the state is memory-only.
RAFT_MTLS_ADDRunsetThe listen address of the mTLS raft peer plane. A voting control plane must set it.
RAFT_ADVERTISE_IP, RAFT_ADVERTISE_HOSTnoneExtra IP and DNS SANs on the mesh leaf certificates.
LOAD_SHED_THRESHOLD0.85The orchestrator’s load-shed threshold.
SNAPSHOT_THRESHOLD, SNAPSHOT_RETAIN1000, 100Raft snapshot cadence and retention, in entries.
WATCHDOG_CPU_THRESHOLD99.5The watchdog’s CPU alarm, in percent.
WATCHDOG_MEMORY_THRESHOLD_MBlibrary defaultThe watchdog’s memory alarm.
WATCHDOG_GOROUTINE_THRESHOLD10000The watchdog’s goroutine alarm.
WATCHDOG_RUNNER_DIAL_TIMEOUT_SECONDS5The runner reachability probe timeout.
CIRCUIT_BREAKER_FAILURE_THRESHOLD5Failures before the breaker opens.
CIRCUIT_BREAKER_RECOVERY_TIMEOUT_SECONDS30How long the breaker stays open before it tries again.
SELF_HEAL_AUTONOMOUSonfalse limits the self-healing reasoner to ENDPOINT_UNHEALTHY events.

Fetch, runners and agents

VariableDefaultPurpose
MANAGED_RUNNER_ENDPOINTface-runner:7102Where the control plane forwards fetches.
REQUIRE_RUNNER_ISOLATIONunsettrue makes an unreachable runner a hard failure instead of a fallback to the control plane’s in-process SQL service.
RUNNER_PORTnoneThe port used when dialling a configured runner endpoint that has none.
RUNNER_ENROLL_ADDRunsetOn a self-hosted runner, the control-plane address it enrols with (see Runners).
RUNNER_ENROLL_TOKENnoneThe service-account token the runner enrols with. It is required when the address is set.
RUNNER_ENROLL_TLStrueTLS with system roots for the enrolment dial. false is refused, and the runner does not enrol.
RUNNER_NAME, RUNNER_VERSIONemptyShown in the control plane’s runner registry.
RUNNER_COMPUTE_POOL, RUNNER_CPU_LIMIT, RUNNER_MEMORY_LIMITnone, 1, 1GInputs to runink runner create for the Snowflake runner type.
FACE_RUNNER_CHANNEL_ADDR, FACE_RUNNER_CHANNEL_LISTENnoneThe address runners in Snowflake or Databricks reach FACE’s runner channel at, and the local listen address. Both must be set for those deploy targets to be offered.
FACE_RUNNER_CA_DIRnoneThe runner CA (ca.crt, ca.key), read by path.
FACE_RUNNER_IMAGE_ARCHIVE, FACE_RUNNER_IMAGE_ARCHIVE_SHA256, FACE_RUNNER_IMAGE_REF, FACE_RUNNER_BINARY, FACE_RUNNER_BINARY_SHA256noneThe runner artifacts an on-premises deploy copies, each with its SHA-256.
FACE_RUNNER_GCP_IMAGEnoneThe default digest-pinned runner image in your registry for a Google Cloud deploy.
SCHEDULE_EXECUTE_FETCHofftrue makes briefs run real fetches. When it is off, a brief’s trigger logs a dry run. The deployment sets it through spec.scheduleExecuteFetch.
FETCH_CACHE_TTL10mHow long a fetch result may be reused. 0 disables reuse.
FACE_RALPH_JUDGEMENTonControls judging of proposed actions.
FACE_RALPH_JUDGE_MAX_ACTIONSset by FACEBounds the judging cost per answer.
FACE_RALPH_JUDGE_REVISIONSset by FACEBounds the revision rounds after dissent.
FACE_FAST_JUDGEMENTonThe local fast-path judge in front of the model.
FACE_A2A_PEERSnoneAgent-to-agent delegation peers, separated by commas, spaces or newlines.
A2A_ORGANIZATIONfalls back to RUNINK_ORG, then TIDE_GITHUB_ORGThe organization published on the agent card.
FACE_AGENT_EMAIL_ALLOWLISTnoneThe addresses or @domain suffixes the audit trail treats as approved recipients for agent e-mail. Agent e-mail is always drafted for a person to send, whatever this list or the decision-authority policy says.
FACE_OUTPUT_URL_HOSTSnoneHosts an agent answer may link to, comma separated; a subdomain matches its parent. Every other link is removed from the answer, and images and embedded frames are always removed.
FACE_GUARD_MODELnoneThe model-router token of the guard tier, usually guard. Configures the guard model that screens text in the agent loop.
FACE_PROMPT_GUARDonControls the guard model configured by FACE_GUARD_MODEL. The OpenBias rules and the output constraints always apply.
FACE_AUTHORITY_POLICY_FILEembedded defaultA decision-authority policy (versioned JSON) replacing the shipped one. Unreadable or invalid = every approval refused.
FACE_OPERATIONAL_LENS_FILEembedded defaultAn operational lens (versioned JSON: areas, signals and the column-name terms that show them, and the decisions each needs) replacing the shipped one for the Operational coverage section on Ask. Unreadable or invalid = no lens, and the page says so.
ENFORCE_AUDIT_LOGGING, REQUIRE_HITL, DATABASE_ENGINE, PRIMARY_SERVER_URLunsetReported into every compliance analysis as the controls in place.

Connectors, vision and integrations

VariableDefaultPurpose
FACE_BROWSER_CONCURRENCY12The number of headless-Chrome renders run at once. Lower it on small hosts.
FACE_MAP_TILE_URLunsetAn optional raster basemap for the twins map: an XYZ tile URL template ({z}, {x} and {y} all required) on your own tile service. Set, it is drawn instead of FACE’s built-in vector basemap. FACE never falls back to a public tile server. A template missing a placeholder is ignored, with a warning in the log.
FACE_MAP_BASEMAP_PMTILES/app/basemap/basemap.pmtilesFACE’s built-in basemap: a PMTiles archive of OpenStreetMap vector tiles that ships in the server image and is served by the server itself at /api/map/basemap/{z}/{x}/{y}.mvt. Point it at a larger archive (for example a regional extract with street-level zooms) to give the map more detail. Missing, the map draws plain ground and says that no basemap is installed.
FACE_MAP_GLYPHS_DIR/app/basemap/glyphsThe glyphs the map draws place names with (<font name>/<start>-<end>.pbf), served at /api/map/glyphs/. Without them the map shows places as dots.
FACE_MAP_TILE_ATTRIBUTION© OpenStreetMap contributorsThe credit shown on the map whenever the basemap is. Tiles built from OpenStreetMap data must carry it under the ODbL, self-hosted included; set your own if your tiles come from other data.
FACE_MAX_DOWNLOAD_BYTES2 GiBThe largest file a connector downloads.
SCRAPER_USER_AGENTMozilla/5.0 (compatible; RuninkFACE/1.0)The user agent for crawling.
VISION_DETECT_INSTANCESofftrue adds a per-instance detection turn per camera frame. It roughly doubles the vision time per frame.
VISION_MAX_EDGE448Frames are downscaled to this longest edge. 0 disables downscaling. Keep it below 1000.
VISION_SWEEP_FRAMES2The number of frames sampled per camera ingest (at most 8).
VISION_RESOLVE_ITEMSofftrue resolves detected objects against the item catalogue.
OCR_VISION_FALLBACKofftrue lets OCR fall back to the vision model.
GOOGLE_CREDENTIALS_JSONsecretA Google Workspace service-account JSON. The stored integration config is used when it is empty.
TWILIO_ACCOUNT_SID, TWILIO_AUTH_TOKEN, TWILIO_FROM_NUMBER, TWILIO_WHATSAPP_FROMnoneVoice, SMS and WhatsApp. FACE refuses to send by name when they are missing.
PUBLIC_URLlocalhost:8080The public host used for Twilio media-stream callbacks.
FACE_VERIFY_MAIL_FROMnoneThe envelope sender for the e-mail deliverability probe in the company checks (VerifyContact). It must be an address this deployment owns. Unset, the e-mail check records NOT_RUN with that reason, and no contact can become VERIFIED.
CALLING_AGENT_URLnoneThe shared calling agent that WebRTC signalling is forwarded to. When it is unset, the call RPC answers Unavailable.

Frontend

PORT (default 8080), WEB_DIR (/web) and BACKEND_ADDR. See Deployment model.

Removed and renamed names

Since the TIDE rename (2026-10-19) FACE reads only the TIDE_* names above (TIDE_PLATFORM_ADMIN, TIDE_ROUTER_IDENTITY_KEY, TIDE_ROUTER_BILLING_KEY_FILE, TIDE_ROUTER_BILLING_KID, TIDE_HEALTH_URL, TIDE_HEALTH_TOKEN, TIDE_GITHUB_ORG, TIDE_ENVELOPE_KEK). The platform renders them; there is no fallback to the pre-TIDE spelling.

These are listed to help an upgrade. FACE ignores every one of them, except OBJECTSTORE_USE_SSL, which it refuses. Remove them from your resources.

NameStatus
REMOTE_LLAMA_URL, LLAMA_REMOTE_URLRenamed to INFERENCE_REMOTE_URL. A manifest that sets only the old name has no inference endpoint, and the process refuses to start.
LLAMA_API_KEYRenamed to INFERENCE_API_KEY.
MESH_MTLS, ENABLE_ALTSRemoved. Mesh mTLS is unconditional.
REQUIRE_MTLS_INGRESSRemoved.
OBJECTSTORE_USE_SSLRemoved, and refused if it is set. Put the scheme in OBJECTSTORE_ENDPOINT.
GOOGLE_API_KEYRead by no code. Maps keys are entered per connection in Settings › Integrations.