SDK Parity

Where each Pilot Protocol SDK stands today and what is planned to close the gap. Three official SDKs ship under one brand: Python, Node.js, and Swift (iOS/macOS). They all speak the same wire protocol, but they do not yet expose the same public API surface.

Status at a glance

Node.js — package pilotprotocol on npm. Coverage: feature complete. TypeScript types, using support, Buffer I/O. The reference surface alongside Python.
Python — package pilotprotocol on PyPI. Coverage: feature complete. Full type hints (py.typed), context managers, snake_case naming.
Swift — package sdk-swift (SwiftPM). Coverage: core trust + datagrams. Embedded daemon (XCFramework, no separate process). Streams, networks, managed networks, policy, member tags, high-level services, and most registry admin are not yet exposed.

Summary by feature category

Updated against sdk-node@d02bd00, sdk-python@93584ea, sdk-swift@0d49f87 (audited 2026-05-28). Each row lists support as Node / Python / Swift.

Lifecycle (construct, dispose) — Node yes, Python yes, Swift yes. Each SDK uses its native idiom (using, with, deinit).
Daemon admin — info, health — Node yes, Python yes, Swift yes.
Daemon admin — rotateKey — Node yes, Python yes, Swift no. Planned for Swift.
Trust — initiate handshake, list trusted peers — Node yes, Python yes, Swift yes.
Trust admin — approve/reject/pending/revoke — Node yes, Python yes, Swift no. Planned for Swift.
Trust convenience — waitForTrust — Node no, Python no, Swift yes. Planned for Node and Python. Currently Swift-only.
Datagrams — sendTo, recvFrom — Node yes, Python yes, Swift yes. Swift uses send(to:port:data:) and returns a typed Datagram.
Datagrams — broadcast — Node yes, Python yes, Swift no. Planned for Swift.
Streams — dial, listen, Conn, Listener — Node yes, Python yes, Swift no. Planned for Swift.
Registry admin — hostname / visibility / deregister / tags / webhook — Node yes, Python yes, Swift no. Planned for Swift.
Networks — list/join/leave/members/invite/respond — Node yes, Python yes, Swift no. Planned for Swift.
Managed networks — score/status/rankings/cycle/reconcile — Node yes, Python yes, Swift no. Planned for Swift.
Policy — get/set — Node yes, Python yes, Swift no. Planned for Swift.
Member tags — get/set — Node yes, Python yes, Swift no. Planned for Swift.
High-level services — sendMessage, sendFile, publishEvent, subscribeEvent — Node yes, Python yes, Swift no. Planned for Swift.
FFI loader — findLibrary / loadLibrary — Node yes, Python private, Swift n/a. Intentional: Swift embeds the library in an XCFramework, so no loader is needed.
Typed response structs (Config, StartResult, Datagram, Error) — Node n/a, Python n/a, Swift yes. Swift idiom. Node/Python return untyped Record<string, unknown> / dict[str, Any] from the same RPCs.

What counts as a gap

Naming differences across languages are not gaps. The matrix collapses idiomatic equivalents into a single canonical row:

Constructors: new Driver() (Node), Driver() (Python), and Pilot.start(config) (Swift) are the same operation expressed in each language's idiom.
Cleanup: driver.close(), pilot.stop(), Python's with, and Node's using all resolve to the same teardown call.
Naming convention: Python's send_message maps to Node/Swift's sendMessage — same method, language-appropriate spelling.
Datagram receive: Node and Python return a dict; Swift returns a typed Datagram struct. Both surface the same underlying RPC.

A real gap is an operation that one SDK does not expose at all. Those are the rows marked unintentional — every one has a follow-up ticket to close it.

Full method matrix

The complete row-per-method matrix — including exact signatures and rationale for each gap — is generated by a deterministic script in the main protocol repo (scripts/parity-audit/). Re-running it against newer commits of the three SDKs produces an updated matrix.csv checked in alongside the script.

Roadmap

End-state target: full parity across all three SDKs, except for the three intentional rows above (FFI loader, socket-path default, and Swift's typed response structs).

Swift: the gap-fill work is tracked as a single follow-up ticket covering streams, networks, managed networks, policy, member tags, registry admin, trust admin, and high-level services. Wire protocol support already exists — what is missing is the Swift surface that exposes it.
Node and Python: add waitForTrust(peerId, timeoutMs) — Swift's blocking convenience. Today, Node and Python users have to poll pendingHandshakes.