Migrating from Dwellir

The full six-method gateway surface is present. Existing stubs call the same methods; the final column states our precise behavior:

Reflection is enabled, so grpcurl … list will show the gateway alongside our native packages — see SDKs & protos.

1. 01

   Swap the endpoint

   Replace the Dwellir endpoint with ours wherever the channel is built. TLS over HTTP/2, port 443:

   grpc endpointdiff

   Copy

   −removed: ENDPOINT = "your-current-dwellir-endpoint:443"

   +added: ENDPOINT = "stream.hyperliquidrpc.com:443"
2. 02

   Move auth to x-api-key metadata

   Authentication is an API key sent as gRPC metadata on each call — replace your previous credential mechanism with the x-api-key pair. Keys come from the console; see Authentication.

   client.pydiff

   Copy

   −removed: channel = grpc.secure_channel(ENDPOINT, grpc.ssl_channel_credentials())

   −removed: gateway = HyperliquidL1GatewayStub(channel)

   −removed: fills = gateway.StreamFills(request)

   +added: channel = grpc.secure_channel(ENDPOINT, grpc.ssl_channel_credentials())

   +added: gateway = HyperliquidL1GatewayStub(channel)

   +added: fills = gateway.StreamFills(request, metadata=(("x-api-key", "YOUR_KEY"),))
3. 03

   Check your channel options

   • Message size. StreamOrderbookSnapshots and GetOrderBookSnapshot responses can be large on liquid markets — set max receive message size to at least 256 MB.
   • Reconnects. Streams are long-lived but not eternal; reconnect with backoff. A DATA_LOSS status means your consumer fell behind — reconnect, and use the gateway's point-in-time GetFills / GetBlock (or native replay from height) to close the gap.
   • History addressing. On Ultra, block_height reads and replay are retained for up to 6 hours. timestamp_ms inputs and order-book snapshot methods resolve to current state; historical timestamp snapshots are not promised.

   Copy

   channel = grpc.secure_channel(    "stream.hyperliquidrpc.com:443",    grpc.ssl_channel_credentials(),    options=[        # order-book snapshots can exceed default limits — allow >=256 MB        ("grpc.max_receive_message_length", 268_435_456),    ],)

4. 04

   Validate side by side

   Run both endpoints in parallel before cutting over. Fills and blocks carry their height, so per-block equality is mechanical:

   validate.pypython

   Copy

   # run both providers in parallel and diff fills per block.# uses the hyperliquid_l1_gateway.v2 stubs you already generated.import grpc OLD = "your-current-dwellir-endpoint:443"NEW = "stream.hyperliquidrpc.com:443" def fills(endpoint, metadata=()):    channel = grpc.secure_channel(        endpoint,        grpc.ssl_channel_credentials(),        options=[("grpc.max_receive_message_length", 268_435_456)],    )    gateway = HyperliquidL1GatewayStub(channel)    return gateway.StreamFills(StreamFillsRequest(), metadata=metadata) old = fills(OLD)                                          # existing authnew = fills(NEW, metadata=(("x-api-key", "YOUR_KEY"),))   # ours # align on a common block height, then compare per-block fill setsfor a, b in zip(old, new):    assert a.block_height == b.block_height, (a.block_height, b.block_height)    assert {(f.oid, f.px, f.sz) for f in a.fills} == \           {(f.oid, f.px, f.sz) for f in b.fills}    print("block", a.block_height, "ok:", len(a.fills), "fills") # current snapshot methods are easy to spot-check too:# gateway.GetOrderBookSnapshot(Timestamp()) against both endpoints

   Let it run through a busy period — funding intervals and a volatility burst are the interesting cases — then cut over.

The gateway gets you migrated; the native packages on the same endpoint and key are why you move:

• Attribution-rich trades — the hyperliquid_swaps.v1 flat schema adds trader wallet, realized PnL, fees and rebate flag, maker/taker, and position context to every fill: trades stream.
• Surfaces beyond the gateway — order lifecycle, raw book diffs, full L4 book, TWAPs, and funding/liquidation events.
• Server-side filtering and replay — subscribe by coin and wallet; on Ultra, replay native stream data from a block within the last 6 hours.
• Manual historical exports — request captured order events or book updates for a confirmed range, using the raw live field names.
• RPC on the same key — HyperEVM JSON-RPC and the HyperCore /info API.