Configuration¶

The backend reads environment variables. .env.example at the repository root lists every variable Compose passes; this page groups them.

Core¶

Variable	Default	Meaning
`STUDIO_ENV`	`dev`	`production` refuses the placeholder `JWT_SECRET` and skips the demo account
`JWT_SECRET`	placeholder	Signs session tokens
`JWT_TTL_SECONDS`	`86400`	Session lifetime
`STUDIO_SECRET_KEY`	generated	Fernet key for stored secrets
`STUDIO_SECRET_KEY_FILE`	`~/.superstep/secret.key`	Where a generated key is kept
`DATABASE_URL`	SQLite `backend/studio.db`	SQLAlchemy URL, e.g. `postgresql://…`
`STUDIO_DB`	`backend/studio.db`	SQLite path when `DATABASE_URL` is empty
`REDIS_URL`	empty	Run bus, queue and counters; empty keeps them in memory
`STUDIO_PUBLIC_URL`	empty	Public base URL for links in e-mails, approvals and webhooks
`STUDIO_CORS_ORIGINS`	`*` in dev, none in production	Allowed browser origins
`STUDIO_COOKIE_SECURE`	`1` in production	`Secure` flag on the session cookie
`WEB_PORT`	`8080`	Compose: port of the `web` service

Accounts¶

Variable	Default	Meaning
`STUDIO_ADMIN_EMAILS`	empty	Bootstrap admins
`STUDIO_DEFAULT_ROLE`	`editor`	Role of new accounts (`editor` or `viewer`)
`STUDIO_LOGIN_MAX_FAILURES`	`5`	Failed sign-ins before a lockout
`STUDIO_LOGIN_WINDOW`	`900`	Window they are counted in (s)
`STUDIO_LOGIN_LOCK`	`900`	Lock duration (s)
`STUDIO_PASSWORD_RESET_TTL`	`3600`	Lifetime of a reset link (s)

Models¶

Variable	Default	Meaning
`OPENAI_API_KEY`	empty	Fallback for LLM nodes and embeddings with no credential
`OPENAI_BASE_URL`	empty	Base URL for that fallback
`TYPESAFE_API_KEY`, `TYPESAFE_BASE_URL`	empty	Fallback for Decision nodes
`STUDIO_PRICES`	empty	JSON price overrides, USD per 1M tokens `{"prefix": [in, out]}`
`STUDIO_PRICES_FILE`	empty	The same, from a file

Runs and limits¶

Variable	Default	Meaning
`STUDIO_RECURSION_LIMIT`	`25`	Steps per run
`STUDIO_RUN_TIMEOUT`	`120`	Run duration (s); import runs default to 300
`STUDIO_EVENT_MAX_BYTES`	`262144`	One event
`STUDIO_RUN_MAX_BYTES`	`2097152`	Stored events per run
`STUDIO_INPUT_MAX_BYTES`	`262144`	Run input
`STUDIO_WORKFLOW_MAX_BYTES`	`1048576`	Workflow JSON
`STUDIO_WAIT_TTL`	`604800`	A waiting run expires after this (s); `0` never
`STUDIO_VERSION_KEEP`	`0`	Versions kept per workflow; `0` keeps all
`STUDIO_TEST_CASES_MAX`	`50`	Test cases per workflow
`STUDIO_TEST_PARALLEL`	`4`	Test cases of one run of the tests that run at a time
`STUDIO_ENABLE_DEMO_TOOLS`	`0`	Registers `sleep_demo` for cancel and timeout tests

Workers and the queue¶

Variable	Default	Meaning
`STUDIO_WORKER`	`1`	Run the worker inside the API process
`STUDIO_WORKER_CONCURRENCY`	`2`	Jobs per worker
`STUDIO_JOB_VISIBILITY_TIMEOUT`	`30`	Job lease (s)
`STUDIO_JOB_REAP_EVERY`	`15`	Requeue expired leases every (s)
`STUDIO_RECOVER_ON_START`	`1`	In-memory queue: requeue unfinished runs at start-up (runs across restarts)
`STUDIO_RECOVER_MAX_ATTEMPTS`	`3`	Restarts that may cut one run off before it ends as `RUN_INTERRUPTED`

API triggers and webhooks¶

Variable	Default	Meaning
`STUDIO_TRIGGER_RATE_LIMIT`	`60`	Runs per token per window; `0` off
`STUDIO_TRIGGER_RATE_WINDOW`	`60`	The window (s)
`STUDIO_CHAT_TRIGGER_RATE_LIMIT`	`60`	Chat triggers: runs per sender and trigger per window; `0` off
`STUDIO_PUBLIC_CHAT_RATE_VISITOR`	`10`	Public chat page: messages per visitor per window; `0` off
`STUDIO_PUBLIC_CHAT_RATE_IP`	`30`	Public chat page: messages per client IP per window; `0` off
`STUDIO_PUBLIC_CHAT_RATE_USER`	same as `STUDIO_PUBLIC_CHAT_RATE_IP`	Messages/uploads per external user and chat per window; `0` off
`STUDIO_PUBLIC_CHAT_RATE_READ`	`120`	Public chat page: page loads, chat reads and new chats per client IP per window; `0` off
`STUDIO_PUBLIC_CHAT_RATE_WINDOW`	`60`	The public chat page's window (s). Each chat's own daily cap is in its Chat settings
`STUDIO_CHAT_ATTACHMENT_MAX_MB`	`10`	Chat attachments: the largest file in MB; each chat may only lower it. Above 100, also raise `WEB_MAX_BODY_SIZE` on the `web` image (default `110m`), which the bundled nginx applies to the upload routes only; every other request is limited to 10 MB there
`STUDIO_MAX_BODY_MB`	`10`	The largest request body in MB, except the upload routes (chat attachments, workflow import, knowledge documents), which have their own ceilings; the chat and approval callbacks take at most 1 MB
`STUDIO_CHAT_ATTACHMENT_MAX_FILES`	`5`	Chat attachments: the most files per message; each chat may only lower it
`STUDIO_ATTACHMENT_STORAGE`	`db`	Where chat attachment bytes are kept: `db` (in the database), `fs` (a folder) or `s3` (S3-compatible storage). Files keep the place they were stored in, so a change applies to new files. See Chat: Attachments
`STUDIO_ATTACHMENT_FS_PATH`	`/data/attachments`	The folder of the `fs` backend. In docker compose, `api` and `worker` share it on the `apidata` volume. It must not be or contain the data folder
`STUDIO_ATTACHMENT_S3_BUCKET`	(empty)	The bucket of the `s3` backend. Required for `s3`
`STUDIO_ATTACHMENT_S3_PREFIX`	(empty)	A folder inside the bucket, so several installs can share one bucket. Leave it empty only on a bucket used by nothing else (TASK-0203)
`STUDIO_ATTACHMENT_S3_ENDPOINT`	(empty)	The service URL for S3-compatible storage (R2, B2, MinIO). Empty means AWS S3
`STUDIO_ATTACHMENT_S3_REGION`	(empty)	The bucket's region; `auto` for Cloudflare R2
`STUDIO_ATTACHMENT_S3_PATH_STYLE`	`false`	`true` for services that need path-style URLs, such as MinIO
`STUDIO_ATTACHMENT_S3_MULTIPART_MB`	`8`	Files larger than this many MB are uploaded to S3 in parts of this size (at least 5); a failed upload is aborted, so no incomplete parts are left behind
`STUDIO_ATTACHMENT_S3_ACCESS_KEY_ID`	(empty)	Access key. With both keys empty, the default AWS credential chain is used (instance role, `AWS_*` variables)
`STUDIO_ATTACHMENT_S3_SECRET_ACCESS_KEY`	(empty)	Secret key. Never logged or shown; `/api/health` only says whether the bucket works
`STUDIO_SYNC_WAIT_MAX`	`30`	Longest `sync` wait (s)
`STUDIO_ALLOW_PRIVATE_CALLBACKS`	`0`	Allow callbacks to private addresses
`STUDIO_WEBHOOK_MAX_ATTEMPTS`	`5`	Delivery attempts
`STUDIO_WEBHOOK_BACKOFF_BASE`	`2`	Backoff base (s)
`STUDIO_WEBHOOK_BACKOFF_MAX`	`300`	Backoff cap (s)
`STUDIO_SECRET_GRACE`	`86400`	Old signing secret stays valid after rotation (s)

Tools, e-mail and MCP¶

Variable	Default	Meaning
`STUDIO_HTTP_TOOL_ALLOW_PRIVATE`	`0`	HTTP Tool nodes may call private addresses
`STUDIO_HTTP_TOOL_TIMEOUT`	`30`	HTTP Tool request timeout (s)
`STUDIO_SMTP_ALLOW_PRIVATE`	`0`	Send email nodes may use private SMTP hosts
`STUDIO_EMAIL_NODE_HOURLY_LIMIT`	`100`	Recipients per workflow owner per hour
`STUDIO_MCP_STDIO`	`0`	Allow stdio MCP servers (admins only)
`STUDIO_MCP_ALLOW_PRIVATE`	`0`	Allow HTTP MCP servers on private addresses
`STUDIO_MCP_TIMEOUT`	`30`	MCP call timeout (s)

Code mode¶

Variable	Default	Meaning
`STUDIO_ENABLE_CODE_NODES`	`0`	Turn Code mode on
`STUDIO_CODE_RUNTIME`	`docker`	`docker`, or `subprocess` (runs on host: no Docker limiting its access)
`STUDIO_CODE_IMAGE`	`studio-code:1`	Sandbox image
`STUDIO_CODE_PACKAGES`	the image's	Packages shown in the editor
`STUDIO_CODE_MEMORY`	`256m`	Memory per call
`STUDIO_CODE_CPUS`	`0.5`	CPU per call
`STUDIO_CODE_PIDS`	`64`	Processes per call
`STUDIO_CODE_MAX_SECONDS`	`30`	Cap on a call's time
`STUDIO_CODE_CONCURRENCY`	`4`	Containers at once per process
`STUDIO_CODE_TEST_PER_MINUTE`	`20`	Editor tests per user per minute

Repo import¶

Variable	Default	Meaning
`STUDIO_ENABLE_IMPORT`	`0`	Turn import on
`STUDIO_IMPORT_RUNTIME`	Docker	`subprocess` only for tests (runs on host: no Docker limiting its access)
`STUDIO_IMPORT_DIR`	`~/.superstep/imports`	Job directories
`STUDIO_IMPORT_HOST_DIR`	empty	Host path of that directory under Compose
`STUDIO_IMPORT_IMAGE`	`python:3.12-slim`	Sandbox image
`STUDIO_IMPORT_MEMORY`	`1g`	Memory
`STUDIO_IMPORT_CPUS`	`1`	CPU
`STUDIO_IMPORT_PIDS`	`256`	Processes
`STUDIO_IMPORT_NETWORK`	`bridge`	Network of a run (`bridge` or `none`)
`STUDIO_IMPORT_SECRETS`	empty	Extra env names passed into the sandbox
`STUDIO_INSPECT_TIMEOUT`	`120`	Inspect time (s)
`STUDIO_IMPORT_MAX_AGE_HOURS`	`72`	Cleanup age; `0` disables
`STUDIO_IMPORT_MAX_BYTES`	`209715200`	Largest source tree
`STUDIO_DOCKER_GID`	`10001`	Compose: group id of the Docker socket

Notifications and approvals¶

Variable	Default	Meaning
`SMTP_HOST`	empty	Studio e-mail (notifications, reset links, e-mail approvals); empty turns mail off
`SMTP_PORT`, `SMTP_USER`, `SMTP_PASSWORD`, `SMTP_FROM`	`587`, empty, empty, `studio@localhost`	SMTP settings
`SMTP_STARTTLS`, `SMTP_SSL`	`1`, `0`	Transport security
`STUDIO_NOTIFY_WEBHOOK_URL`, `STUDIO_NOTIFY_WEBHOOK_SECRET`	empty	JSON webhook for waiting and expired runs
`STUDIO_APPROVAL_TTL`	`86400`	Lifetime of approval buttons and links (s)
`STUDIO_IDENTITY_CODE_TTL`	`600`	Verification code lifetime (s)
`STUDIO_IDENTITY_CODE_ATTEMPTS`	`5`	Tries per code
`STUDIO_IDENTITY_CODE_RESEND`	`30`	Resend wait (s)
`SLACK_CLIENT_ID`, `SLACK_CLIENT_SECRET`, `SLACK_SIGNING_SECRET`	empty	Add to Slack
`WHATSAPP_API_URL`	Graph API	WhatsApp Cloud API base
`STUDIO_TRUSTED_PROXIES`	private ranges	Proxy CIDRs for `X-Forwarded-For`
`STUDIO_CALLBACK_RATE_LIMIT_IP`	`60`	Public callbacks per IP per window
`STUDIO_CALLBACK_RATE_LIMIT_CHANNEL`	`300`	Per connection per window
`STUDIO_CALLBACK_RATE_WINDOW`	`60`	The window (s)

Tracing¶

Variable	Default	Meaning
`OTEL_EXPORTER_OTLP_ENDPOINT`	empty	OTLP/HTTP endpoint; empty turns tracing off
`OTEL_EXPORTER_OTLP_HEADERS`	empty	OTLP headers
`OTEL_SERVICE_NAME`	`superstep`	Service name
`STUDIO_TRACE_URL`	empty	Open trace link template with `{trace_id}`
`STUDIO_OTEL_METRICS`	`0`	Also export metrics
`LANGSMITH_API_KEY`, `LANGSMITH_TRACING`	empty	Export to LangSmith
`LANGSMITH_PROJECT`, `LANGSMITH_ENDPOINT`	empty	LangSmith project and endpoint