Environment Variables
The environment variables the CLI reads, grouped by what they configure. This page was rebuilt from the data-product-forge 0.18.1 source. A variable is listed with the behaviour its read site implements; where a variable that earlier versions of this page listed does nothing in 0.18.1, it is named below.
fluid doctor --env # runtime kill switches: current value, source, default
fluid doctor --env prints only the switches doctor knows about (forge copilot stages, cost ceilings, run id, the OpenTofu and federation timeouts, DBT_EXECUTABLE, telemetry). (since 0.31.0, forge-cli #888) Telemetry is not listed. The rest are read where they are used.
How values are parsed
There is no single rule. Some variables accept only 1, some accept 1/true/yes/on, and a few treat any non-empty value as set, including 0. The tables say which where it matters. When in doubt, use 1 to turn something on and unset the variable to turn it off.
Where you set it matters
Since 0.20.0, a project's .env, .env.<environment> and .env.local may hold credentials but not a setting that widens a sandbox, moves where the CLI reads trusted configuration, changes authentication or trust, changes where data and credentials are sent, or chooses which program runs. fluid names what it ignored and runs without it. Export those in your shell, your CI job, or a FLUID_SECRETS_FILE you export — see What a project's .env may set.
(since 0.33.0, forge-cli #1004) A project file can no longer set a name that is not a shell identifier, a name that starts with BASH_FUNC_, BASH_XTRACEFD, ZDOTDIR, FPATH, NULLCMD or READNULLCMD either. (since 0.33.0, forge-cli #975) Nor can it set a name that starts with SPARK_, PYSPARK_, HADOOP_ or YARN_.
CLI behaviour
| Variable | Effect |
|---|---|
FLUID_LOG_LEVEL | Default of --log-level: DEBUG, INFO, WARNING or ERROR. Default INFO. |
FLUID_LOG_FILE | Default of --log-file: also write logs to this file. |
FLUID_QUIET | 1 suppresses banners, the agent progress panel and the capability-warning print. Only the value 1 counts. |
FLUID_NONINTERACTIVE | 1 suppresses the same chrome and skips the interactive fluid forge surfaces. Only the value 1 counts. It does not change whether fluid apply prompts; see Operating in CI. |
FLUID_NO_TUI | 1 silences the agent progress panel. |
FLUID_BANNER | 1, true, yes or on opts in to the roadmap teaser banner. As of 0.18.1 it prints nothing: the teaser's last milestone has passed and the banner expired on 2026-05-07. |
FLUID_BANNER_TODAY | Overrides "today" in the banner's date arithmetic. Test use only. |
FLUID_BUILD_PROFILE | experimental (default) or stable; stable curates the registered command set. |
FLUID_RUN_ID | Pre-seeds the cross-stage run id instead of generating one. Set it in CI to correlate the stages of one pipeline run. |
FLUID_TELEMETRY | 1 enables anonymous usage telemetry, 0 forces it off and overrides the config file. Off by default. (since 0.31.0, forge-cli #888) Removed. The telemetry and its consent are gone, so the variable does nothing. |
DO_NOT_TRACK | When set, all telemetry is off, whatever else is configured. (since 0.31.0, forge-cli #888) Removed. Forge no longer reads it, and fluid doctor --env no longer lists it. telemetry.enabled in ~/.fluid/config.yaml does nothing either. |
FLUID_SHOW_REGISTRATION | 1, true or yes prints the provider and extension registration lines that --log-level DEBUG also prints. |
Project, region and paths
| Variable | Effect |
|---|---|
FLUID_PROVIDER | Default of the global --provider. A binding's platform wins over it when a plan resolves the provider. |
FLUID_PROJECT | Default of the global --project; the GCP provider also reads it as a project fallback. |
FLUID_REGION | Default of the global --region. The flag's own default is europe-west3. |
FLUID_DEFAULT_REGION | A project-wide region for the class-based provider plan path. The order is --region, the contract's region, a binding's location.region, then FLUID_DEFAULT_REGION, then AWS_REGION and AWS_DEFAULT_REGION, then eu-west-1. |
FLUID_ENV | Not a default for --env: validate, bundle, plan and apply ignore it, so pass --env. It is read by fluid generate artifacts (the env baked into scheduled DAGs), by fluid schedule-sync --env, and by the dotenv credential store, which loads .env.<env>. Generated pipelines set it as a build parameter and pass --env "${FLUID_ENV:-dev}" to each stage. |
FLUID_HOME | Base directory for the copilot config.yaml and prices.json, and for artifact paths. Default ~/.fluid. |
FLUID_CONFIG | Path of the copilot config.yaml; wins over FLUID_HOME. |
FLUID_USER_HOME | Overrides the user-global FLUID directory (~/.fluid). |
FLUID_WORKSPACE_ROOT | Overrides the directory that holds the workspace-local .fluid/ state. Default: the working directory. |
FLUID_SECRETS_FILE | Path of a dotenv-style file that is loaded last, over the process environment, when a command hydrates the project's .env files (.env, .env.<env>, .env.local, then this file). Best effort, and it needs python-dotenv. It is not an encrypted file. |
FLUID_UPSTREAM_CONTRACTS | Colon-separated extra directories searched for upstream contracts, next to the contract's own workspace. Used when fluid forge and the dbt source generator look up an upstream's schema, and when an embedded-SQL build binds a consumes[] entry to its upstream's binding. Each root is searched down to four directory levels for contract.fluid.yaml or contract.fluid.json; build output and VCS directories such as .git and node_modules are skipped. See Workspaces. |
Contract loading and the DuckDB sandbox
| Variable | Effect |
|---|---|
FLUID_REF_ROOT | Widens the directory tree a $ref may resolve under (CLI 0.18.0). Ignored, with a ref_root_env_ignored warning, for a contract that lies outside it. See Composing a contract with $ref. |
FLUID_DUCKDB_ALLOWED_DIRS | Directories, separated by os.pathsep, where a contract may also declare inputs and outputs for the local DuckDB engine (CLI 0.18.0). Set by the operator, never by the contract. See the DuckDB sandbox. |
FLUID_LOCAL_DUCKDB_PATH | Database file for the legacy fluid_build.contract_tests module. Default :memory:. |
Apply, state and OpenTofu
OpenTofu state explains how these variables choose the state key.
| Variable | Effect |
|---|---|
FLUID_STATE_BACKEND | Default of --state-backend for fluid apply, fluid diff and fluid verify --state-drift: s3://<bucket>[/<key>] or gcs://<bucket>[/<prefix>] (see Remote state). A spec that names only a bucket gives every contract its own key (see below); the flag does not. An empty --state-backend "" forces local state for one run. |
FLUID_STATE_ENCRYPTION | (since 0.31.0, forge-cli #905) Opt-in encryption of the OpenTofu state and saved plan: aws-kms://..., gcp-kms://..., openbao://<transit mount>/<key> or pbkdf2+<secret reference>. A protected name that a project's .env cannot set. Needs OpenTofu 1.13 or newer. See State and plan encryption. |
TF_CLI_ARGS, TF_CLI_ARGS_<command>, TF_VAR_*, TF_ENCRYPTION, TF_INPUT, TF_REATTACH_PROVIDERS | (since 0.31.0, forge-cli #905) Never reach tofu: forge runs it with an allowlist of the environment, and a tofu_env_dropped warning names what it dropped. See What tofu gets from the environment. |
FLUID_TOFU_TIMEOUT_SECONDS | Wall-clock cap per tofu invocation. Default 1800. A timed-out call reports exit code 124. (since 0.27.0, forge-cli #787) tofu is interrupted first, so it releases its state lock, and killed only if it is still running 30 seconds later. |
FLUID_DATABRICKS_RUN_TIMEOUT_SECONDS | (since 0.30.0, forge-cli #861) How long fluid apply --mode amend-and-build or replace-and-build waits for a Databricks job run, in whole seconds, before it cancels the run and fails the build. Unset, an hour, or the job's own timeouts and retries plus 15 minutes when that is longer. A value that is not a whole number of seconds above 0, in at most nine digits, is not used, and the apply says so. fluid doctor --env lists it, and a project .env may set it. See Running the job from fluid apply. (since 0.30.0, forge-cli #872) It also caps the wait for a Lakeflow pipeline update, an hour when unset, and at the cap the pipeline is stopped (see Running it from fluid apply). |
FLUID_TOFU_S3_LOCKFILE | (since 0.27.0, forge-cli #787) 0, false, no or off stops fluid apply from asking tofu init to lock S3 state (-backend-config=use_lockfile=true, OpenTofu 1.10 or newer). Set it for a role that cannot write <key>.tflock, in the environment of the job: a project .env cannot set it (see project .env trust). Unset, or any other value, keeps the lock on. See Locking. |
FLUID_OPENTOFU_VERSION | Pins the OpenTofu version that fluid apply --ensure-opentofu installs when tofu is missing. (since 0.31.0, forge-cli #899) The default pin is 1.13.9. |
FLUID_REDSHIFT_WORKGROUP, FLUID_REDSHIFT_DATABASE, FLUID_REDSHIFT_SQL | Set by the generated AWS module for its Redshift Serverless statements. They are outputs of the generator, not settings. |
SNOWFLAKE_AUTHENTICATOR, SNOWFLAKE_WORKLOAD_IDENTITY_PROVIDER, SNOWFLAKE_TOKEN_FILE_PATH | (since 0.31.0, forge-cli #905) WORKLOAD_IDENTITY, then OIDC, AWS, AZURE or GCP, and for OIDC the file that holds the CI job's ID token. Connects without a user or a password. See Workload identity federation. |
SNOWFLAKE_ORGANIZATION_NAME, SNOWFLAKE_ACCOUNT_NAME | The account identity that the generated module's Snowflake provider 2.x expects. fluid derives them from a SNOWFLAKE_ACCOUNT in <org>-<account> form when they are not set. |
SNOWFLAKE_ACCOUNT | Blanked in the environment handed to tofu once a complete v2 identity resolves, because provider 2.x errors on seeing it. Unchanged as a source credential; see Credential resolver. |
With a bucket-only FLUID_STATE_BACKEND, the remote state key is fluid/<id>/<provider>/terraform.tfstate (GCS prefix fluid/<id>/<provider>), so one contract applied to two clouds through overlays keeps two states. A spec that names a key is used as written. The first apply after upgrading to 0.17.0 moves state from the old key fluid/<id>/terraform.tfstate when it holds this provider's resources; the errors that can follow are in Production troubleshooting.
export FLUID_STATE_BACKEND=s3://<your-state-bucket> # bucket only: per-contract, per-provider keys
fluid apply runtime/plan.json --bundle runtime/bundle.tgz --env aws --yes
Spark on AWS (preview)
(since 0.31.0, forge-cli #904, forge-cli #912, forge-cli #918, forge-cli #919, forge-cli #924, forge-cli #927, forge-cli #934) Spark to Iceberg on AWS says what each of these protects. An unset allowlist allows nothing. Every *_ALLOWLIST and *_URL setting, and FLUID_SPARK_MAVEN_REPOSITORY, is read from the operator's environment: a project .env cannot set it.
| Variable | Effect |
|---|---|
FLUID_SPARK_TARGET_ALLOWLIST | The existing runtimes a contract may name (emr-serverless:<applicationId>, emr-ec2:<clusterId>, emr-eks:<virtualClusterId>) and the MSK clusters it may read (msk:<cluster ARN>). (since 0.33.0, forge-cli #975) It also lists the topic a stream's dead letters go to: msk-write:<cluster ARN>/<topic>, or kafka-write:<host:port>/<topic> for each broker a bootstrapServers cluster lists. See Listing the topic. |
FLUID_SPARK_EXECUTION_ROLE_ALLOWLIST | The roles parameters.spark.executionRoleArn may name: a role ARN, or an ARN prefix ending in /. |
FLUID_SPARK_CONF_ALLOWLIST | sparkConf keys beyond the built-in tuning keys: a key, or a prefix ending in *. A key that loads code, sets credentials or moves the catalog or checkpoint is refused even when listed. |
FLUID_SPARK_CONTRACT_SQL_ALLOWLIST | The builds whose own properties.sql a Spark job may run: <product>, <product>/<build>, or <product>/<build>@sha256:<digest>. (since 0.33.0, forge-cli #990) fluid validate reads it too, and warns only about a build it does not approve: see Contract SQL and fluid validate. |
FLUID_SPARK_WAIT_SECONDS | How long a reconcile waits for a cancelled run to stop and a started one to leave its pending states. Default 900. (since 0.34.0, forge-cli #1008) It is also how long fluid apply waits, for each run, for the runs and the application that a destroy would leave going: see When a build leaves the contract. A value that is not a positive number is ignored. |
FLUID_SPARK_MAVEN_REPOSITORY | One https Maven mirror that replaces Confluent's repository in spark.jars.repositories. A URL with a user or password, a query, a comma, plain http or a link-local host is refused. |
FLUID_SPARK_GLUE_JARS_URL | s3://<bucket>/<prefix>/: where the platform keeps the pinned jars of a Glue Confluent Avro build. |
FLUID_SPARK_SUBMIT_EXECUTABLE | The spark-submit a build of parameters.spark.runtime: local starts: an absolute path to an executable outside the project, or a name on PATH. Default: spark-submit on PATH. (since 0.33.0, forge-cli #975) Running a build on this machine. fluid doctor --env lists it. |
FLUID_SPARK_SUBMIT_PROPERTIES_FILE | An absolute path to the Spark properties file that every local run gets as --properties-file: the catalog fluid of the output, the S3 store, their credentials and the master. Default: none. (since 0.33.0, forge-cli #975) Running a build on this machine. fluid doctor --env lists it. |
FLUID_SPARK_LOCAL_ROOT | Where local runs read their job config and runner: s3://<bucket>[/<prefix>] (through the AWS SDK's AWS_ENDPOINT_URL_S3) or an absolute directory. Default: .fluid/spark-artifacts beside the contract. (since 0.33.0, forge-cli #975) Running a build on this machine. fluid doctor --env lists it. |
FLUID_SPARK_LOCAL_NAMESPACE_ALLOWLIST | The namespaces a local run may create and write tables in, as <warehouse>/<namespace> (fluid/<namespace> when the binding names no warehouse), comma-separated, whole levels. An unset list allows none. (since 0.33.0, forge-cli #975) The table the job creates. fluid doctor --env lists it. |
FLUID_AIRFLOW_ALLOW_UNMAPPED | 1 renders a task whose action the provider cannot map as a log-only task, instead of refusing. (since 0.31.0, forge-cli #923) The same as --allow-unmapped on fluid generate schedule. |
SCHEMA_REGISTRY_REST_ENDPOINT, SCHEMA_REGISTRY_ID, SCHEMA_REGISTRY_API_KEY, SCHEMA_REGISTRY_API_SECRET | (since 0.31.0, forge-cli #913) The registry OpenTofu registers a produced Confluent subject in, and its credential. The first two cannot come from a project .env. See Schema registry. |
Managed Flink (preview)
(since 0.27.0, forge-cli #784) Managed Flink to Iceberg says what each of these protects. Each takes a comma-separated list. An unset list allows nothing, and a project .env cannot set any of them.
| Variable | Effect |
|---|---|
FLUID_FLINK_KUBE_CONTEXT_ALLOWLIST | The kube contexts a managed Flink build may name in deployment.context. |
FLUID_FLINK_NAMESPACE_ALLOWLIST | The namespaces it may deploy into. |
FLUID_FLINK_IMAGE_ALLOWLIST | The images it may run: an exact repository, without :tag and @sha256:, or a prefix ending in /. |
FLUID_FLINK_SERVICE_ACCOUNT_ALLOWLIST | The service accounts the job may run as: name, in any listed namespace, or namespace/name. |
FLUID_FLINK_SECRET_ALLOWLIST | The Kubernetes Secrets deployment.secretRefs may mount, in the same two forms. |
FLUID_FLINK_CONTRACT_SQL_ALLOWLIST | The builds whose own properties.sql fluid apply may run: <product>, <product>/<build>, or <product>/<build>@sha256:<digest> for that SQL only. The refusal prints the pinned entry. (since 0.33.0, forge-cli #990) fluid validate reads it too: see A build's own SQL. |
FLUID_ACQUISITION_DEPLOYMENT_HOST_ALLOWLIST | For managed Flink: every hostname the job contacts (the catalog and an OAuth2 token endpoint at apply and verify; at fluid apply also an s3.endpoint, the Kafka brokers and the schema registry), and every public IP address. Link-local addresses are always refused. Also read by acquisition deployments and by the Kafka Connect and embedded Debezium Iceberg sinks, where it lists a catalog or token-endpoint hostname the worker resolves itself (Operator allowlists). |
FLUID_ACQUISITION_SECRET_REF_ALLOWLIST | A catalogAuth variable, as NAME@host, for each host that receives it or a token minted from it. Read by managed Flink and by the Kafka Connect and embedded Debezium Iceberg sinks (Operator allowlists); acquisition sources use it for their secretRef. A bare NAME allows every host. (since 0.31.0, forge-cli #893) It now covers a secretRef of every scheme, with more entry forms: see Secret references. |
The two FLUID_ACQUISITION_* variables are not new: acquisition deployments read them too. (since 0.27.0, forge-cli #818) The Kafka Connect and embedded Debezium Iceberg sinks read them as well, and a project .env cannot set either.
Lakekeeper grants (fluid policy apply)
(since 0.33.0, forge-cli #993) The operator sets these in the real environment of the job that runs fluid policy apply: a project .env cannot set them. See Lakekeeper tables.
| Variable | Effect |
|---|---|
FLUID_LAKEKEEPER_CREDENTIAL_REF | A secret reference to client_id:client_secret, for an OAuth2 client-credentials token. Needs FLUID_LAKEKEEPER_OAUTH2_SERVER_URI and FLUID_LAKEKEEPER_SCOPE. |
FLUID_LAKEKEEPER_OAUTH2_SERVER_URI | The token endpoint, an https:// URL. |
FLUID_LAKEKEEPER_SCOPE | The scope to ask the token endpoint for. |
FLUID_LAKEKEEPER_TOKEN_REF | A secret reference to a bearer token, used as it is. Set this or FLUID_LAKEKEEPER_CREDENTIAL_REF, not both. |
FLUID_LAKEKEEPER_NAMESPACE_ALLOWLIST | The namespaces whose tables stage 8 may change: comma-separated <warehouse>/<namespace prefix> entries, matched on whole dotted levels. Unset allows none, and --mode enforce then refuses every Lakekeeper table. |
Command Center
The Command Center says what each of these sends.
fluid publish and fluid apply read different variables. Publishing to the Command Center uses the catalog configuration; fluid apply reuses that configuration to report each run, and falls back to the reporter's own variables.
| Variable | Read by | Effect |
|---|---|---|
FLUID_CC_ENDPOINT (or FLUID_CATALOG_FLUID_CC_URL) | fluid publish, and fluid apply run reports | Command Center base URL for the fluid-command-center catalog target. |
FLUID_API_KEY (or FLUID_CATALOG_FLUID_CC_TOKEN) | same | API key sent as X-API-Key. |
FLUID_BEARER_TOKEN | same | Bearer token, when the catalog's auth type is bearer. |
FLUID_CC_ORG_ID | same | Organization id sent as X-Organization-Id. A set but blank value is an error (cc_organization_id_blank). Without it, the organization comes from organization_id or an organization slug in the FLUID config, else from the credential's only organization. |
FLUID_COMMAND_CENTER_URL, FLUID_COMMAND_CENTER_API_KEY | fluid market and fluid marketplace detection, the observability reporter, and as the fallback for fluid apply run reports | They do not change where fluid publish sends a contract. |
FLUID_COMMAND_CENTER_ENABLED | the reporter and fluid apply run reports | 0, false, no or off turns off apply run reports. |
FLUID_COMMAND_CENTER_TIMEOUT | the reporter | Request timeout in seconds. Default 5. |
FLUID_COMMAND_CENTER_HOST_ALLOWLIST | detection, the reporter, apply run reports | Comma-separated host suffixes allowed even when they resolve to a private address. Loopback is always allowed. |
FLUID_DISABLE_CC_DETECTION | fluid market, fluid marketplace | true, 1 or yes skips the Command Center auto-detection. |
FLUID_API_URL | fluid marketplace | Explicit marketplace API URL; wins over detection. |
The Command Center reporter reads a CA bundle from REQUESTS_CA_BUNDLE, or CURL_CA_BUNDLE when that is unset. (from 0.27.0) If the path cannot be loaded, the reporter is disabled for the run with a warning; TLS verification is never relaxed. See Reporting to the Command Center.
fluid apply registers each run at POST /api/v1/executions and closes it with a PATCH, best effort: an unreachable Command Center costs a warning line and at most the timeout, never the exit code. With no organization, nothing is sent. It reports the product id, contract version, hash, env, provider, mode, state location, resource addresses, change counts, timings and the builds it ran. It does not send secrets, headers or tofu output. Under Jenkins (JENKINS_URL set) the runner is tagged jenkins.
Generated pipelines and Airflow workers
The CLI itself does not read these; the generated pipeline and the generated DAG do.
| Variable | Read by | Effect |
|---|---|---|
FLUID_PACKAGE_SPEC | the generated Jenkins stage 0 | Package spec installed into the workspace venv. Defaults to the generating CLI version with the extras the contract needs, for example data-product-forge[local]==0.35.0. |
FLUID_PIP_INDEX_URL, FLUID_PIP_EXTRA_INDEX_URL | the generated install step | Primary and extra pip index. Blank means PyPI. pip takes the highest version across both, so for private packages use one mirror that proxies PyPI in FLUID_PIP_INDEX_URL and leave the extra index empty; see Operating in CI. |
FLUID_ALLOW_PRERELEASE | the generated install step | true adds pip --pre. |
FLUID_CONFIG_PATH | set by generated pipelines | ./fluid_config. Nothing in the CLI reads it. |
FLUID_PROJECT_DIR | the generated Airflow DAG, on the worker | Directory that holds the product checkout. Required. |
FLUID_BIN | the generated Airflow DAG | The fluid executable on the worker. Default fluid. |
FLUID_DAG_ENV_PASSTHROUGH | the generated Airflow DAG | Space-separated extra variable names handed to the fluid apply process. A name that starts with AIRFLOW never passes, so the worker's Airflow configuration and connection variables stay out of fluid. |
FLUID_DAG_CONTRACT, FLUID_DAG_ENV, FLUID_DAG_BUILD_ID, FLUID_DAG_CONTRACT_ENV | the generated Airflow DAG | Set by the DAG's task from constants in the file; do not set them on the worker. |
See Airflow for which variables reach fluid apply on the worker and Operating in CI for the pipeline parameters.
Build runners, masking and lineage
| Variable | Effect |
|---|---|
FLUID_PII_HASH_SECRET | Salt for strategy: hash in policy.privacy.masking. At least 16 bytes. An unset salt refuses the build rather than hashing without one. |
FLUID_PII_TOKENIZATION_KEY | HMAC key for strategy: tokenize and the tokenize_pii hook. At least 32 bytes for masking. Unset, the masking build is refused. |
FLUID_PII_ENCRYPTION_SECRET_KEY | Base64 of a 16, 24 or 32 byte AES key for strategy: encrypt. |
FLUID_RUNNER_HOST_OVERRIDE | Host the runners use to reach containers they start. Wins over TESTCONTAINERS_HOST_OVERRIDE. |
FLUID_DBT_FORWARD_ENV | Extra environment variable names or prefixes the dbt runner forwards. |
DBT_EXECUTABLE | The dbt command the runner and the --dbt-validate and --dbt-tests-key auto probes resolve first: an absolute path, a bare name on PATH, or a multi-token wrapper. Falls back to dbt on PATH, then the active venv. |
FLUID_DBT_TESTS_KEY | auto (default), tests or data_tests: which YAML key generated dbt tests attach under. The --dbt-tests-key flag wins. |
DBT_DOCKER_IMAGE | A prebuilt image for the dbt container fallback. (since 0.34.0, forge-cli #1009) It must be pinned by digest (name[:tag]@sha256:<digest>), and a tag is refused. See Which dbt fluid apply runs. |
DBT_BOOTSTRAP_IMAGE | Base image the dbt container fallback installs dbt into. Default python:3.12-slim. (since 0.34.0, forge-cli #1009) The default is python:3.12-slim pinned by digest, and an override must be pinned by digest too. |
DBT_ADAPTER_PACKAGE | Overrides the pip package that fallback installs. Without it the install is dbt-<adapter><2 plus dbt-core<2; setting it is how you opt in to dbt 2. (since 0.34.0, forge-cli #1009) Without it the install is the adapter's pinned package (dbt-athena-community==1.11.1 for athena) plus dbt-core==1.11.17. It takes one or more exact requirements (name==version) separated by spaces or commas, and forge adds dbt-core==1.11.17 unless the list names dbt-core: name it with the version you want to run another dbt-core. A range is refused. |
FLUID_SOURCE_DATABASE, FLUID_SOURCE_SCHEMA | Read through dbt's env_var() in generated sources.yml: when set, they supply a source's database and schema before the dbt target's own. |
FLUID_ROLLBACK_KEEP_LAST_N | Per-product retention of .fluid/rollback-state.json. Default 20. |
FLUID_OPENLINEAGE_URL (or the standard OPENLINEAGE_URL) | Base URL of an OpenLineage receiver. Without one, no lineage is emitted. Emitting needs the lineage extra. (since 0.28.0, forge-cli #837) |
FLUID_OPENLINEAGE_ENDPOINT | Path appended to the URL. |
FLUID_OPENLINEAGE_API_KEY (or OPENLINEAGE_API_KEY) | Bearer key for the receiver. |
FLUID_OPENLINEAGE_TIMEOUT_SECONDS | Request timeout. Default 5. |
FLUID_OPENLINEAGE_ALLOW_PRIVATE | Defaults to allowing private receiver addresses; false restores the public-only gate. Link-local and metadata addresses stay blocked. |
FLUID_WEBHOOK_HOST_ALLOWLIST | Comma-separated host suffixes the build webhook alerter may call; see Network safety. |
Masking is applied only on the DuckDB acquisition landing path. See Production troubleshooting for the errors.
Verify, import and cloud APIs
| Variable | Effect |
|---|---|
FLUID_ATHENA_OUTPUT_LOCATION | Default of fluid verify --athena-output-location: where Athena writes the row-count result. A workgroup that enforces its own location wins. |
FLUID_ATHENA_WORKGROUP | Default of the Athena workgroup for the row-count query. Default primary. |
FLUID_ATHENA_TIMEOUT_SECONDS | Query timeout. Default 300. |
FLUID_IMPORT_AIRBYTE_URL | Airbyte API base URL for fluid import airbyte. The importer has no default endpoint: without --server-url or this variable it refuses before building a client. |
GOOGLE_PROJECT, GOOGLE_CLOUD_PROJECT, GCLOUD_PROJECT, CLOUDSDK_CORE_PROJECT | Read in this order for the BigQuery load project when the binding names none. |
BIGQUERY_EMULATOR_HOST | Points the BigQuery client at an emulator. fluid then uses anonymous credentials, so no token reaches the emulator. |
AWS_ENDPOINT_URL_S3, AWS_ENDPOINT_URL | Point DuckDB's S3 access at a non-AWS endpoint, such as a local object-store emulator; AWS_ENDPOINT_URL_S3 wins when both are set, and AWS_IGNORE_CONFIGURED_ENDPOINT_URLS=true turns the override off. A value that is not an http(s) URL naming a host, or that carries a user and password, is ignored with a warning. When DuckDB cannot be pointed at a valid override the build fails with ObjectStoreEndpointError rather than write to AWS. |
FLUID_GCP_PROJECT | Counted by the fluid forge welcome scan as a sign that GCP is configured. |
A BigQuery binding that declares no location.region gets a load job with no location, so the job runs where the table is. Pin location.region in the binding when the data's location matters.
Federation and network safety
Federated upstreams says how the check uses these.
| Variable | Effect |
|---|---|
FLUID_FEDERATION_TIMEOUT_SECONDS | Per-git-operation cap when fluid apply fetches a federated upstream digest. Default 30. A non-numeric, zero, negative or non-finite value is ignored with a federation_timeout_invalid warning. |
FLUID_FEDERATION_HOST_ALLOWLIST | Comma-separated host suffixes a federation endpoint may use even when they resolve to a private address. |
FLUID_ALLOW_METADATA_SERVICE | 1 lets the metadata-source credential resolver fall back to the cloud metadata service. Same as --allow-metadata-service. See Credential resolver. |
FLUID_AUDIT_ROOT | Redirects where fluid mcp output-port writes audit events (default ~/.fluid/store/audit/), for example to a path a SIEM forwards. When the contract's agentPolicy.auditRequired is true and this is unset, the gateway prints a notice that names the default. |
Catalog publish-side registrars
See the catalog overview.
| Variable | Effect |
|---|---|
FLUID_CATALOG_DATAHUB_URL, FLUID_CATALOG_DATAHUB_TOKEN | DataHub endpoint and token. |
FLUID_CATALOG_DATAHUB_SPEC_BASE_URL | Base URL for spec source documents. |
FLUID_CATALOG_DATAHUB_MAX_RETRIES | Retry count for DataHub calls. |
FLUID_CATALOG_DMM_URL, FLUID_CATALOG_DMM_TOKEN | Data Mesh Manager endpoint and API key. |
FLUID_CATALOG_OPENMETADATA_URL, FLUID_CATALOG_OPENMETADATA_TOKEN | OpenMetadata endpoint and bearer token. |
Cost and budget gates
See cost tracking.
| Variable | Effect |
|---|---|
FLUID_COST_LIMIT_USD | Per-run LLM cost ceiling in USD. |
FLUID_COST_LIMIT_USD_PER_RUN | Per-run ceiling shown in the progress prefix. |
FLUID_COST_LIMIT_USD_PER_PRODUCT | Per-product LLM cost ceiling, enforced in the agent coordinator. |
FLUID_STAGE_BUDGET_<STAGE>_S | Wall-clock budget in seconds for one forge pipeline stage, for example FLUID_STAGE_BUDGET_LOGICAL_S. Stage names are lowercase in the code (logical, builder, readme, transformation, validator). |
FLUID_MISSION_TIMEOUT_SECONDS | (since 0.31.0, forge-cli #881) Removed. fluid mission is gone. Overrides a mission's wall-clock budget. |
FLUID_PRICES_JSON | Path of a price override file. Without it: $FLUID_HOME/prices.json, then ~/.fluid/prices.json. |
FLUID_TOKEN_COUNTER | chars switches token counting to a character estimate. Any other value keeps the default counter. |
LLM provider and routing
See LLM providers and the LiteLLM backend. (since 0.29.0, forge-cli #850) With nothing here or in a provider key set, a non-interactive run picks a provider itself; see When nothing is configured. (since 0.29.0, forge-cli #852) When nothing selects a provider and none is found, forge stops with copilot_no_llm_provider. (since 0.29.0, forge-cli #854) A provider saved with fluid ai setup comes before that pick, in a run without a terminal too; see Which provider a run uses.
| Variable | Effect |
|---|---|
FLUID_LLM_PROVIDER | Default provider. |
FLUID_LLM_MODEL | Default model. (since 0.29.0, forge-cli #854) Like --llm-model, it passes over a coding agent, whether forge would pick it or fluid ai setup saved it. |
FLUID_LLM_API_KEY | (since 0.29.0, forge-cli #853) API key for a provider named with --llm-provider, FLUID_LLM_PROVIDER or FLUID_FORGE_AGENT; there it wins over the provider's own key variable, such as OPENAI_API_KEY. A provider forge picks itself gets its own key variable, then its keyring key, and a warning says FLUID_LLM_API_KEY was not used; see Which key a provider gets. Through 0.28.0 the key went to whatever provider the run used, ahead of its own key variable. The key does not choose the provider. (since 0.29.0, forge-cli #852) With none selected, forge stops with copilot_no_llm_provider; through 0.28.0 it sent the key to gemini. |
FLUID_LLM_ENDPOINT | Overrides the model endpoint. (since 0.29.0, forge-cli #850) The routing model's calls use it too, unless FLUID_LLM_ROUTING_ENDPOINT names another. (since 0.29.0, forge-cli #852) For Ollama, forge's model pick and model check ask it too, and each --tiered stage calls it. (since 0.29.0, forge-cli #854) It also passes over a coding agent that fluid ai setup saved. |
FLUID_LLM_TIMEOUT_SECONDS | Per-call timeout. Default 120. |
FLUID_LLM_STREAMING | Default on; 0, false, no or off disables streaming. |
FLUID_LLM_ROUTING_ENDPOINT, FLUID_LLM_ROUTING_MODEL | Optional routing endpoint and model. |
FLUID_LLM_MODEL_PREFLIGHT | Runs a model-availability check before the first call. |
FLUID_LLM_BACKEND | litellm routes every LLM call through the unified backend. |
FLUID_LLM_FALLBACK_CHAIN | (since 0.31.0, forge-cli #881) Removed. The fallback chain is gone. Fallback model chain for the router. |
FLUID_LITELLM_MODEL_PREFIX | Prefix prepended to the model name for providers LiteLLM names differently. |
FLUID_TIERED | Uses per-stage model tiers when an LLM is configured. |
FLUID_OPENAI_STRICT_SCHEMA | (since 0.31.0, forge-cli #881) Removed. No longer read. 1, true, yes or on hardens the response schema for OpenAI strict mode. |
FLUID_GEMINI_RESPONSE_SCHEMA | 1 sends the response schema to Gemini. Off by default. |
FLUID_FORGE_AGENT | Selects a keyless coding-agent provider (claude-code, codex, cursor, kiro). |
FLUID_FORGE_AGENT_MODE, FLUID_FORGE_AGENT_TIMEOUT_SECONDS, FLUID_FORGE_AGENT_CWD | (since 0.31.0, forge-cli #881) agentic is the default drive mode, and envelope makes one call through the agent, with no tools. In 0.29.0: Drive mode (envelope or agentic), per-invocation wall-clock cap, and working directory for the agent CLI. |
FLUID_AGENT_WEB_TOOLS | (since 0.31.0, forge-cli #881) Removed. The agent loop and its web tools are gone. 1 exposes the web_search and web_fetch agent tools. |
FLUID_WEB_SEARCH_PROVIDER | (since 0.31.0, forge-cli #881) Removed. The web tools are gone. Forces the web-search backend. |
Provider-specific keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, AWS_* for Bedrock, the Vertex settings) follow the underlying SDKs and are read by LiteLLM.
Forge copilot
(since 0.31.0, forge-cli #881) Removed, and no longer read by main: FLUID_INTERVIEW_LEGACY, FLUID_AGENT_COMPACT_AFTER, FLUID_COMPACTION_STRATEGY, FLUID_PROMPT_OVERLAYS, FLUID_PROMPT_PROFILE, FLUID_OVERLAY_STRICT, FLUID_OVERLAY_PUBLIC_KEYS, FLUID_FORGE_WATCH_INTERVAL and FLUID_FORGE_WATCH_DEBOUNCE, in the last rows of the table below.
| Variable | Effect |
|---|---|
FLUID_FORGE_NO_PICKER | Skips the mode menu on bare fluid forge. Any non-empty value counts. |
FLUID_FORGE_NO_PREVIEW | Skips the pre-write preview panel. Any non-empty value counts. (since 0.31.0, forge-cli #881) The panel is removed; the variable skips AI mode's Keep this contract question. |
FLUID_FORGE_NO_STREAMING_PREVIEW | (since 0.31.0, forge-cli #881) Removed. The streaming contract preview is gone. Disables the live contract-growth panel. Any non-empty value counts. |
FLUID_FORGE_NO_WELCOME | Suppresses the welcome scan. Any non-empty value counts. |
FLUID_FORGE_PICKER_ALWAYS | Shows the picker even for return users. |
FLUID_FORGE_AUTO_CI | 0, false, no or off disables the CI scaffold that a forge run otherwise emits. |
FLUID_FORGE_OFFLINE | 1 forces the local, no-network guided interview. Same as --offline. |
FLUID_FORGE_AUTO_RESUME, FLUID_FORGE_NO_PRUNE_HINT | (since 0.31.0, forge-cli #881) Removed. Resume is gone. 1 resumes an interrupted run without asking, and hides the prune hint. |
FLUID_FORGE_LEGACY_COPILOT | (since 0.31.0, forge-cli #881) Removed. The copilot runtimes are gone. Uses the linear runtime instead of the staged copilot. Any non-empty value counts. |
FLUID_FORGE_STAGED_COPILOT, FLUID_FORGE_STAGED_TOOL_LOOP | (since 0.31.0, forge-cli #881) Removed. The copilot runtimes are gone. 1 selects the experimental staged copilot and tool loop. |
FLUID_FORGE_DRIFT_GUARD | (since 0.31.0, forge-cli #881) Removed. Enables the semantic-drift guard. Opt-in. |
FLUID_FORGE_DB_TOOLS, FLUID_FORGE_DB_URI | (since 0.31.0, forge-cli #881) Removed. The agent tools are gone. Enable the row-capped fetch_sample_rows agent tool, and the connection URI it uses. |
FLUID_FORGE_TOOL_SEARCH | (since 0.31.0, forge-cli #881) Removed. The agent loop is gone. Enables lazy tool-search deferral in the agent loop. |
FLUID_GITHUB_MCP, FLUID_SNOWFLAKE_MCP, FLUID_DBT_MCP | (since 0.31.0, forge-cli #881) Removed. Hosted MCP delegation is gone. Enable delegation to the hosted GitHub, Snowflake and dbt MCP servers. |
FLUID_HOSTED_MCP_TIMEOUT_SECONDS | (since 0.31.0, forge-cli #881) Removed. Hosted MCP delegation is gone. Per-operation timeout for hosted MCP calls. |
FLUID_COPILOT_AGENT_LOOP | (since 0.31.0, forge-cli #881) The agent loop is removed, but the variable was still read: set to a true value, it stopped forge from picking a saved or detected coding agent. (since 0.31.0, forge-cli #888) The variable does nothing, and a saved or detected coding agent is picked whatever it says. In 0.29.0: 1, true, yes or on selects the multi-turn agent loop. Same as --agent-loop. (since 0.29.0, forge-cli #850) The loop also keeps a non-interactive run from picking a coding agent; see When nothing is configured. (since 0.29.0, forge-cli #854) It passes over a coding agent that fluid ai setup saved, too. |
FLUID_COPILOT_JUDGE, FLUID_COPILOT_ENRICHMENT, FLUID_JUDGE_SELF_CRITIQUE, FLUID_COPILOT_SELF_EVAL, FLUID_COPILOT_CHECKPOINT, FLUID_COPILOT_PII_CLASSIFIER | (since 0.31.0, forge-cli #881) FLUID_COPILOT_JUDGE, FLUID_JUDGE_SELF_CRITIQUE and FLUID_COPILOT_SELF_EVAL are removed (the judge and self-evaluation are gone). The others in this row are still read. On by default; 0, false, no or off skips that stage. |
FLUID_COPILOT_POST_SYNTHESIS, FLUID_COPILOT_EPISODIC_MEMORY | On by default; only the value 0 turns them off. |
FLUID_COPILOT_SEMANTIC_MEMORY | (since 0.31.0, forge-cli #881) Removed. Semantic memory is gone. Opts in to semantic-memory writes and lookups. |
FLUID_COPILOT_FORCE_INTERVIEW | (since 0.31.0, forge-cli #881) Removed. The interview is gone. Forces the interview even when it would otherwise be skipped. Any non-empty value counts. |
FLUID_COPILOT_PARALLEL_PHYSICAL | On by default; 0, false, no or off runs the physical-modeling stages one at a time. |
FLUID_INTERVIEW_LEGACY | (since 0.31.0, forge-cli #881) Removed. 1 reverts to the legacy bootstrap interview. |
FLUID_AGENT_COMPACT_AFTER | (since 0.31.0, forge-cli #881) Removed. Compact the agent context after this many turns. Default 6. |
FLUID_COMPACTION_STRATEGY | (since 0.31.0, forge-cli #881) Removed. truncate, summarize or hybrid. |
FLUID_PROMPT_OVERLAYS, FLUID_PROMPT_PROFILE, FLUID_OVERLAY_STRICT, FLUID_OVERLAY_PUBLIC_KEYS | (since 0.31.0, forge-cli #881) Removed. Prompt overlays and profile for fluid forge; strict mode and the keys that verify signed overlays. |
FLUID_FORGE_WATCH_INTERVAL, FLUID_FORGE_WATCH_DEBOUNCE | (since 0.31.0, forge-cli #881) Removed. Poll interval and debounce, in seconds, for fluid forge watch mode. |
MCP authoring server
(since 0.31.0, forge-cli #884) Read by fluid mcp serve; see Large results.
| Variable | Effect |
|---|---|
FLUID_MCP_INLINE_TOKEN_LIMIT | The largest tool result, in estimated tokens, that the server sends inline. Default 6000; 0 sends every result inline. A larger result comes back as a fluid://artifacts/ resource. A value that is not a whole number, or is below 0, logs a warning and the default applies. Set it in the environment the server starts with, such as the env block of the client's MCP config: fluid mcp serve does not read a project .env. |
MCP output port
Read by fluid mcp output-port; see MCP. (since 0.31.0, forge-cli #879) fluid mcp serve --transport http reads FLUID_MCP_AUTH_MODE, FLUID_MCP_AUTH_TOKEN and the FLUID_MCP_JWT_* variables too.
| Variable | Effect |
|---|---|
FLUID_MCP_AUTH_MODE | shared-token (default), jwt or none. |
FLUID_MCP_AUTH_TOKEN | The shared token. |
FLUID_MCP_JWT_ISSUER, FLUID_MCP_JWT_AUDIENCE, FLUID_MCP_JWT_JWKS_URL | Required for jwt mode. |
FLUID_MCP_JWT_ALGORITHMS, FLUID_MCP_JWT_CLAIM_MAPPING | Accepted algorithms, and extra claim mappings merged over the defaults. |
FLUID_MCP_RATE_LIMIT, FLUID_MCP_RATE_WINDOW_SECONDS | Sliding-window rate limit. Default 60 calls per 60 seconds; 0 disables it. |
FLUID_MCP_MAX_CONCURRENCY | In-flight tool calls per gateway process. Default 8; 0 disables. |
FLUID_MCP_CIRCUIT_THRESHOLD, FLUID_MCP_CIRCUIT_WINDOW_SECONDS, FLUID_MCP_CIRCUIT_COOLDOWN_SECONDS | Circuit breaker. Defaults 5 failures, 60 s window, 30 s cooldown. |
FLUID_MCP_AUDIT_WEBHOOK_URL, FLUID_MCP_AUDIT_WEBHOOK_HEADER_AUTH, FLUID_MCP_AUDIT_WEBHOOK_TIMEOUT_SECONDS | Forward audit events to a webhook. |
FLUID_AUDIT_MAX_AGE_DAYS, FLUID_AUDIT_MAX_TOTAL_MB | Audit rotation. Defaults 30 days and 256 MB. |
Store, secrets and encryption
| Variable | Effect |
|---|---|
FLUID_STORE_BACKEND | (since 0.31.0, forge-cli #881) The postgres and vector backends are removed: in 0.31.0 those values log a warning and use the file store. Memory store for the forge copilot: file (default), sqlite, postgres or vector; 0, none, null or disabled turns persistence off. |
FLUID_STORE_ROOT | Root of the file backend. Default ~/.fluid/store. |
FLUID_STORE_PATH | Database file of the sqlite backend. |
FLUID_STORE_DSN | (since 0.31.0, forge-cli #881) Removed. Main no longer reads it. Connection string of the postgres backend. Required for it. |
FLUID_STORE_VECTOR_BACKING | (since 0.31.0, forge-cli #881) Removed. Main no longer reads it. Backing store of the vector backend: file (default), sqlite or postgres. |
FLUID_SECRETS_INMEMORY | 1 keeps fluid secrets values in memory instead of the OS keychain. Tests only. |
FLUID_ENCRYPTION_KEY, FLUID_ENCRYPTION_PASSPHRASE | (since 0.31.0, forge-cli #893) Removed. The Fernet credential file store (~/.fluid/credentials.enc) is gone, and nothing reads these. Keep a credential in the keyring, a secret store or the environment. Fernet key, or a passphrase it is derived from, for the encrypted credential store. |
FLUID_ALLOW_PLAINTEXT_AI_SECRETS | Legacy plaintext fallback for AI provider secrets. (since 0.29.0, forge-cli #852) It also lets fluid ai setup save a user:password@ in an address, such as an Ollama host, to ~/.fluid/ai_config.json; without it the address is saved without them. |
FLUID_ALLOW_PLAINTEXT_SOURCE_SECRETS | Legacy plaintext secrets in sources.yaml; the file must also be mode 600. |
Secret references
(since 0.31.0, forge-cli #893) The operator's policy for secretRef values, the Vault and OpenBao settings, and the two handoffs. Every name here is operator-only, so a project's .env cannot set it, except BAO_TOKEN and VAULT_TOKEN. Those two are credentials, and a project .env may supply either one to a command that loads it, when that variable is not already set in your real environment (even to an empty value) or in a FLUID_SECRETS_FILE you exported. BAO_TOKEN is read first, so a project BAO_TOKEN beats an exported VAULT_TOKEN: see Which token is used. The others come from the real environment, your CI job or a FLUID_SECRETS_FILE you export. fluid secrets explains each one.
| Variable | Effect |
|---|---|
FLUID_SECRET_SCHEMES | Comma-separated schemes that may resolve. Default env,file,vault,aws-sm. A pack's scheme resolves only when listed. See FLUID_SECRET_SCHEMES. |
FLUID_SECRET_PREFIXES | Comma-separated <scheme>://<locator-prefix> entries a contract's references must sit under. Default: file under /run/secrets/, no limit for the other schemes. See FLUID_SECRET_PREFIXES. |
FLUID_SECRET_STAGES | Comma-separated stages that may fetch a value a contract names. Default apply. validate, plan, bundle and diff never fetch, whatever is listed. See FLUID_SECRET_STAGES. |
FLUID_ACQUISITION_SECRET_REF_ALLOWLIST | The hosts a value may be sent to, for every scheme: NAME or NAME@host for env://NAME, <reference>@<host> or <reference>@*, and <prefix>*@<host>. Default empty: only a loopback host. See the allowlist. |
FLUID_RUN_SECRETS_FILE | Path of a mode 0600 JSON file of {"<reference>": "<value>"} that a runner wrote. While it is set, references resolve from it and no store is asked, and a reference it does not list is refused. See the run-secrets file. |
FLUID_CREDENTIAL_STORE | A store the provider credential chain asks after the OS keyring: vault://<mount>/<prefix> or aws-sm://<prefix>. Each credential is read from <prefix>/<provider>#<key>. Unset means no store is asked. See Provider credentials from a store. |
BAO_ADDR, VAULT_ADDR | Address of the OpenBao or Vault store. BAO_ADDR wins when both are set. https, or http to a loopback address only. See Vault and OpenBao. |
BAO_TOKEN, VAULT_TOKEN | The store token. BAO_TOKEN wins when both are set. It is used after FLUID_VAULT_WRAPPED_TOKEN and FLUID_VAULT_JWT_ROLE, and before ~/.vault-token. Not operator-only: a project .env can set either one that is not already set in the real environment, and a project BAO_TOKEN then beats your VAULT_TOKEN. See Which token is used. |
BAO_NAMESPACE, VAULT_NAMESPACE | The store's namespace, sent as X-Vault-Namespace. BAO_NAMESPACE wins when both are set. |
BAO_CACERT, VAULT_CACERT | A PEM CA bundle that verifies the store's certificate. BAO_CACERT wins when both are set. TLS verification is never switched off. |
FLUID_VAULT_JWT_ROLE | The role of a JWT login, for CI. Setting it selects that login over BAO_TOKEN. See JWT login for CI. |
FLUID_VAULT_JWT, FLUID_VAULT_JWT_FILE | The JWT for that login, inline or in a file of at most 64 KiB. The inline value wins. |
FLUID_VAULT_JWT_MOUNT | The JWT auth mount. Default jwt. |
FLUID_VAULT_WRAPPED_TOKEN | A response-wrapped token that a runner hands to one run. It is unwrapped once and wins over every other token source. See A response-wrapped token. |
Marketplace
| Variable | Effect |
|---|---|
FLUID_PUBLIC_REGISTRY | Public registry URL used when FLUID_MARKETPLACE_FALLBACK=public. |
FLUID_MARKETPLACE_FALLBACK | local (default), public or none: what fluid marketplace uses when no Command Center marketplace is available. |
FLUID_MARKET_CACHE_TTL | Cache lifetime in minutes. |
FLUID_MARKET_DEFAULT_LIMIT | Default page size for fluid market. |
FLUID_MARKET_MIN_QUALITY | Minimum quality score for listed products. |
FLUID_MARKET_TIMEOUT | Request timeout in seconds. |
Plugins and providers
| Variable | Effect |
|---|---|
FLUID_PROVIDERS | Comma-separated provider modules to load instead of the default set. |
FLUID_PLUGIN_STRICT_COMPAT | 1, true or yes fails on a plugin compatibility mismatch instead of warning. |
FLUID_PLUGINS_ALLOWLIST, FLUID_PLUGINS_BLOCKLIST | Allow or block plugins by name. Names match case-insensitively, with - equal to _. An allowlist entry of the form <distribution>:<name> (acme-fluid-aws:aws) is the operator grant that lets an IaC plugin take a built-in platform name, and restricts nothing else. See plugins. |
Variables with no effect in 0.18.1
Earlier versions of this page listed the variables below. In 0.18.1 nothing reads them, or what reads them is not part of the running CLI, so setting them changes nothing. No consumer was found for these in the 0.18.1 source.
| Variable | What to use instead |
|---|---|
FLUID_SAFE_MODE, and the global --safe-mode flag | Neither refuses a network call. See Network safety for the checks that do exist. |
FLUID_AUTO_CONFIRM | fluid apply --yes. fluid apply asks for confirmation only on an interactive terminal. |
FLUID_DRY_RUN, FLUID_DEBUG, FLUID_TRACE, FLUID_PROFILING, FLUID_RICH_OUTPUT | --dry-run, --debug, --profile, --no-color on the command. |
FLUID_MAX_FILE_SIZE_MB, FLUID_PARALLEL_OPERATIONS, FLUID_CACHE_ENABLED, FLUID_CACHE_TTL_SECONDS, FLUID_TIMEOUT_SECONDS | None. |
FLUID_LOG_FORMAT | Read into the config object, but FLUID_LOG_FORMAT=text did not change the stderr format when measured on 0.18.1. |
FLUID_DIAG_OUT_DIR | fluid doctor --out-dir. doctor sets this variable for the diagnostic script it starts; it does not read it. |
FLUID_CONFIG_PATH | None. Generated pipelines set it; the CLI does not read it. |
FLUID_DOMAIN, FLUID_LAYER_PROPERTY_ID, FLUID_PRODUCT_TYPE_PROPERTY_ID, FLUID_TIME_GRAINS, FLUID_LLM_TEMPERATURE, FLUID_MAJORS | None. These are not read anywhere in the package. |
FLUID_EXTRA_PIP_SPECS | Named in a note that fluid generate ci prints for engines it cannot install, but no generated pipeline reads it. Put the extras in --fluid-package-spec. |
FLUID_MARKET_URL | Named in the market_discovery_failed suggestion, but not read. Use FLUID_API_URL, FLUID_PUBLIC_REGISTRY and FLUID_MARKETPLACE_FALLBACK. |
Python 3.10 and the litellm cap
litellm 1.98.0 added a module that imports NotRequired straight from typing, which landed there in Python 3.11, so importing litellm raised ImportError on 3.10 while its metadata still declared 3.10 support. pyproject.toml therefore splits the pin on an environment marker: litellm>=1.83.7,<2 on 3.11 and newer, litellm>=1.83.7,<1.98 below it. The >=1.83.7 floor is the same on both branches. The second line goes away once upstream guards that import.
See also
- Operating in CI: the pipeline parameters and the variables a runner needs
- Production troubleshooting: errors and what to do
- LiteLLM backend: LLM variables in context
- Cost tracking: the LLM cost ceilings in context. The acquisition
cost.budgetis described under source-aligned acquisition - Network safety: SSRF allowlists in context
- Catalog overview: publish-side variables in context
fluid generate iac: IaC engine variables in context