A small, documented wire protocol — no linking, no client library, nothing to build against.
Reamer Stream ingests OHLCV data under a ticker name — from a vendor API, an in-house pipeline, or a CSV export — and serves it back over reamer-stream, a Unix-socket server any tool can talk to without linking C++. A Databento connector, an Alpaca connector, an in-house research engine, and reamer_py's own backtest engine all reach that same data the same way.
Ingest once, under a ticker name, from a CSV or from already-structured bars. Query it back — by ticker, by time range, unbounded — from any connected caller, or several at once, all reading the same files.
reamer_py's own backtest engine · anything that opens a socketOne binary, one protocol, one data directory — every connected caller reads and writes the same files, so an ingest from one connection is immediately queryable from any other.
Download reamer-stream, run it, connect to the socket — no client library, no symlinking, no separate pieces to place. PROTOCOL.md is the entire interop contract.
A Unix domain socket, not a TCP port — no listening address, no network hop for a packet to traverse, and no TLS handshake to pay for on every query.
A bespoke, fully documented wire protocol — write a client in Rust, Go, Java, or anything that can open a socket, without waiting for an official binding or linking C++ yourself.
Any socket connection can subscribe to a named stream; any other can publish to it. Subscribers get pushed the record the moment it's published — no polling, no relationship to what's stored on disk unless you choose to reuse the same name.
OhlcvBar is sampling-agnostic — time bars, tick bars, volume bars, dollar bars, or imbalance bars all fit the same 64-byte record. Reamer Stream ingests whatever your own pipeline already builds; constructing the bars themselves stays your pipeline's job.
0600 socket permissions by default, plus an optional shared-secret token for same-user process isolation — a stated, documented trust model for same-machine IPC, not an afterthought bolted onto a network protocol that assumed a different threat model.
Every query and ingest is logged — ticker, byte count, success or failure — for compliance review and debugging, on by default, not an opt-in you have to remember to turn on.
Every number below comes from an actual run against a live reamer-stream subprocess on this session's own machine (Intel i5-8250U, 16GB — the same machine reamer_py's own benchmark page uses). See the full benchmark page for methodology.
Socket round-trip: 19.1µs average, ~52,500 queries/sec for a real AF_UNIX round trip through reamer-stream — wire encoding and epoll dispatch fully included, the actual product experience, not an in-process shortcut.
reamer-stream is single-threaded by design (see the Concurrency semantics section of the protocol docs), so these numbers measure one client, not concurrent multi-client load.
import reamer_py
# Ingest into Reamer Stream under a ticker name (auto-detects CSV format)
reamer_py.load_csv("aapl.csv", "AAPL")
# Query it back — from this same process, or from any other tool
# connected to reamer-stream over the same data directory
result = reamer_py.run_backtest(data=["AAPL"], strategy=MyStrategy())A plain socket connection, stdlib only — the same path any other language's client takes:
import socket, struct
def pack_str(s):
b = s.encode(); return struct.pack(">H", len(b)) + b
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
sock.connect("/tmp/reamer-stream.sock")
# QueryHistorical: type 0x00, ticker, no time bounds
payload = b"\x00" + pack_str("AAPL") + b"\x00" + b"\x00"*8 + b"\x00" + b"\x00"*8
sock.sendall(struct.pack(">BI", 1, len(payload)) + payload)
# HistoricalResponse comes back the same way — full framing in PROTOCOL.mdreamer_py or another), a vendor API connector, or a research notebook querying historical bars directly without going through a full backtest run.
A single Reamer Labs license activates a machine — reamer-stream checks for that active license at startup (activate it with reamer-stream activate <key>, or python -m reamer_py activate on a machine that also runs reamer_py — either one covers both products). Contact us for pricing and multi-seat/invoicing terms.