pm-ws

A venue-agnostic prediction-market data-feed daemon: it connects to venue WebSocket feeds (Limitless first), keeps authoritative binary-market order books, and publishes latest state plus level mutations to local consumers at ultra-low latency over redundant connections.

• docs/design.md — behavior and invariants
• docs/limitless.md — venue facts and live etiquette
• docs/roadmap.md — the ordered work list
• AGENTS.md — project rules

Operate

cargo install --git https://github.com/codebuster22/pm-ws --locked pm-ws

or, from a clone, cargo install --path . --locked. Either installs pmwsd (the daemon), pmwsctl (control CLI: add/remove/status), and pmws-run (single-market live tool).

The quickest look is one live market, no config file:

pmws-run --market <market-slug> --seconds 30 --print-book

Market identity is venue-native; docs/limitless.md covers market discovery and live etiquette.

A minimal config, e.g. pmwsd.toml:

control_socket = "/tmp/pmwsd.sock"
endpoint = "wss://ws.limitless.exchange/socket.io/?EIO=4&transport=websocket"
markets = ["btc-up-or-down-5-min-<epoch>"]
metrics_listen = "127.0.0.1:9090"

Run the daemon, then inspect it:

pmwsd --config pmwsd.toml
pmwsctl --socket /tmp/pmwsd.sock status
curl 127.0.0.1:9090/metrics

pmwsctl add <slugs...> / remove <slugs...> change the operator-pinned market set at runtime. See docs/design.md for behavior and invariants, docs/limitless.md for venue facts and live etiquette.

Consume

The daemon maps each shard's books into a shared-memory segment file under the [delivery] directory (default /tmp, one pmws-<instance>-<shard>.seg per shard). Consumers attach read-only and follow latest state plus the mutation stream, with every price and quantity an exact scaled decimal, never a float.

Python (bindings/python/pmws.py, standard library only):

import pmws
with pmws.Segment("/tmp/pmws-<instance>-<shard>.seg") as segment:
    market = segment.resolve("limitless", "slug", "<market-slug>")
    state, stream = market.attach()
    print(state.best("Bid"), state.best("Ask"))

TypeScript: bindings/node/pmws.ts, the same Segment/Market/EventStream shape. Rust: embed the crate, or read a segment directly (examples/reader.rs). Runnable strategy-shaped consumers ship as examples:

python3 examples/bbo.py --segment <path> --market <slug> --events
node examples/bbo.ts --segment <path> --market <slug>

Both bindings load the daemon's compiled library from target/release, so consumers run from a clone built with cargo build --release.

Verify

./check

Runs cargo fmt --check, cargo clippy --workspace --all-targets --locked -- -D warnings, and cargo test --workspace --locked. Before a release or a dependency change, also run:

cargo deny check && cargo audit

Before a release, also regenerate third-party notices:

cargo about generate about.hbs -o THIRD-PARTY-NOTICES.md

Pinned tools: cargo-deny 0.20.2, cargo-audit 0.22.2, cargo-about 0.9.2. Toolchain is pinned in rust-toolchain.toml.

Benchmark

bench/make_table.sh produces the ship benchmark deterministically, with no venue traffic. bench/reports/ holds the checked-in results, each with the command that produced it.

Licence

MIT, Chaain Labs — see LICENSE. Third-party crate notices, including full licence texts, are in THIRD-PARTY-NOTICES.md.