Binding configuration

OpenOrange reads bindings from the same Wrangler configuration as wrangler dev. Each binding is one of three kinds:

• Runs as is: OpenOrange implements the Cloudflare API. You set nothing.
• Needs a service: OpenOrange implements the Cloudflare API on a service that you supply, such as your SMTP server. You set one variable that points to it.
• Not supported: the binding loads, but Orange has no backend for it. The Worker loads, and env.NAME exists. Each method call throws an error that names the binding and suggests an alternative. Analytics Engine is the exception: writeDataPoint() accepts data points and drops them, as wrangler dev does.

orange doctor reports each binding of your project as one of these kinds. It does not list variables (vars, .dev.vars, and .env), because they need nothing on Rivet.

Where a binding stores data depends on the command. Under orange dev, Rivet Actors run on a local Rivet Engine, and data stays on your machine. Under orange deploy, they run on your Rivet cluster. A service binding sends data to the service that you set.

Matrix

Binding	Wrangler key	Kind	Runs on	Notes
Variables	vars	Runs as is	The Worker process	Strings and JSON values.
Local secrets	.dev.vars, .env	Runs as is	The Worker process	Loaded like wrangler dev. See Variables and secrets.
Secrets Store	secrets_store_secrets	Runs as is	The Worker process	Local values come from .dev.vars or the process environment. See Secrets Store.
Version metadata	version_metadata	Runs as is	The Worker process	See Version metadata.
unsafe_eval	unsafe.bindings	Runs as is	The Worker process
Static assets	assets	Runs as is	Local files	See Static assets.
Durable Objects	durable_objects	Runs as is	One Rivet Actor for each object	Partial. Same-Worker classes from new_sqlite_classes and new_classes migrations. See Durable Objects.
Durable Objects from another Worker	durable_objects with script_name	Runs as is	The other Worker’s Rivet Actors	Both Workers run in one orange dev. See Multiple Workers.
Service bindings to this Worker	services	Runs as is	The Worker process	fetch(), JS RPC, named entrypoints, props, and ctx.exports.
Service bindings to another Worker	services	Runs as is	The other Worker’s process	Partial. Run both Workers in one orange dev with repeated -c. connect() and WebSocket upgrades throw. See Multiple Workers.
Dispatch namespaces	dispatch_namespaces	Runs as is	The Workers of the orange dev session	Partial. See Multiple Workers.
KV	kv_namespaces	Runs as is	One Rivet Actor for each namespace	See KV.
D1	d1_databases	Runs as is	One Rivet Actor for each database	See D1.
Rate Limiting	ratelimits, unsafe.bindings	Runs as is	One Rivet Actor for each namespace	See Rate Limiting.
R2	r2_buckets	Runs as is	One Rivet Actor for each bucket, or your S3-compatible store	Set ORANGE_R2_<BUCKET>_S3_* to use Amazon S3, Cloudflare R2, Google Cloud Storage, or Backblaze B2. See R2.
Queues	queues.producers, queues.consumers	Runs as is	One Rivet Actor for each queue	See Queues.
Workflows	workflows	Runs as is	Cloudflare’s own Workflows engine, with one Rivet Actor for each instance	Partial. See Workflows.
Tail Workers	tail_consumers	Runs as is	Rivet Durable Streams	Partial. See Tail Workers.
Worker Loader	worker_loaders	Runs as is	One Node.js process for each dynamic Worker, in an agentOS VM	Partial. See Worker Loader.
Workers AI	ai	Needs a service	An OpenAI-compatible provider	Set OPENROUTER_API_KEY, or ORANGE_AI_BASE_URL for OpenAI, Groq, Together AI, or Fireworks AI. See Workers AI.
Send Email	send_email	Needs a service	Your SMTP server	Set ORANGE_EMAIL_SMTP_URL to Amazon SES, Resend, Postmark, SendGrid, or Mailgun. Without it, the binding throws. See Email.
Browser Rendering	browser	Needs a service	A hosted browser or your Chrome, through its DevTools Protocol endpoint	Set ORANGE_BROWSER_CDP_URL to Browserbase, Browserless, or Steel. Without it, the binding throws. See Browser Rendering.
Hyperdrive	hyperdrive	Needs a service	Your database, with a direct connection	Set the local connection string, as for wrangler dev, such as PlanetScale, Amazon RDS, Neon, or Supabase. No pooling or query cache. See Hyperdrive.
Analytics Engine	analytics_engine_datasets	Not supported	Nothing	writeDataPoint() accepts data points and drops them, as wrangler dev does.
Vectorize	vectorize	Not supported	Nothing	See Other bindings.
Images, Media, Stream	images, media, stream	Not supported	Nothing	See Other bindings.
Pipelines	pipelines	Not supported	Nothing
AI Search	ai_search	Not supported	Nothing
mTLS certificates	mtls_certificates	Not supported	Nothing
VPC	vpc_networks, vpc_services	Not supported	Nothing	Use fetch() to reach the host network directly.
Legacy blobs	text_blobs, data_blobs, wasm_modules	Not supported	Nothing	Import the file from the Worker source instead.
Other binding types		Not supported	Nothing	Includes other unsafe binding types.

• Configuration that changes how the Worker runs, not what env contains, still stops orange dev with Unsupported Wrangler configuration.
• Examples are streaming_tail_consumers and triggers.events.
• orange dev accepts containers with a warning and runs @cloudflare/sandbox classes on a sandbox provider, such as E2B, Daytona, or Freestyle (see Sandbox).
• orange dev runs triggers.crons (see Cron Triggers) and delivers queues.consumers to the Worker’s queue() handler (see Queues).

Bindings that are not supported

orange dev marks each binding that is not supported in its binding table at startup, with what to use instead on the next line:

  Binding        Resource            Runs on
! env.IMAGES     Images              stub · calls throw
                                     See the OpenOrange bindings documentation for supported bindings.

The binding is a stub, and it is safe to pass around. typeof env.IMAGES is "object", reading a property returns a function, and awaiting the stub does not call it. Calling any method throws an Error with the same message:

await env.IMAGES.info(stream);
// Error: Images binding "IMAGES" is not supported by OpenOrange. ...

What to use instead

Most bindings that are not supported have a replacement. Most replacements use fetch() to an external service, or a Durable Object, which OpenOrange runs as a Rivet Actor with SQLite storage and alarms.

Other bindings

OpenOrange does not support these bindings. Change the Worker to call one of these services instead:

Binding	Service to use instead	How the Worker reaches it
Vectorize	pgvector on Neon or Supabase	A Hyperdrive binding and a Postgres driver
Vectorize	Qdrant Cloud	fetch() to https://<cluster host>:6333, with the api-key header
Vectorize	Pinecone	fetch() to the index host, with the Api-Key header
Vectorize, a few thousand vectors	SQLite in a Durable Object	Store the embeddings, and rank them by cosine similarity in JavaScript
Images	Cloudinary	fetch() to its upload and transformation URLs
Images	imgproxy that you run	fetch() to its processing URLs
Analytics Engine	Your analytics store	fetch() to its ingest API

Variables and secrets

orange dev resolves variables with Wrangler’s own wrangler dev logic:

1. It starts with vars from the Wrangler configuration.
2. It loads .dev.vars.<env> or .dev.vars next to the configuration file.
3. If neither file exists, it loads .env, .env.local, .env.<env>, and .env.<env>.local. Set CLOUDFLARE_INCLUDE_PROCESS_ENV=true to also load the process environment.
4. Loaded values become string bindings that replace vars with the same name.
5. If secrets.required is set, only the listed keys load, and orange dev warns about missing keys.

A .dev.vars entry named after another binding, such as a Secrets Store binding, stays that binding’s local value and does not replace it.

Secret values never become part of the version metadata ID.