sääl | wiring up the core

← back sääl | wiring up the core Jan 9, 2026

Quick update on sääl’s engineering progress. I’ve been focused on getting the wire-level core in place first, so there’s a solid, deterministic foundation before anything network-y or storage-y lands.

So far

The repo now has a saal-core crate that’s basically the minimal spec-compliant kernel:

Canonical CBOR encoding for unsigned objects so every node hashes the same bytes.
ObjectId as BLAKE3-256 over canonical CBOR of the unsigned envelope.
Ed25519 signatures over the ObjectId, plus verification helpers.
Typed payloads for the v0.1 wire spec: Profile, Post, Reaction, Follow, MediaBlob, and LogEntry.
Tests for deterministic IDs, sign/verify round-trips, and canonical CBOR stability.

There’s also a draft wire spec (wire-spec-v0.1.md) in the repo that mirrors what the code is doing. The idea is to keep spec and code aligned while the protocol is still fluid.

This is intentionally boring in a good way:

CBOR + canonical ordering gives compact payloads and deterministic hashes. No "same data, different bytes" surprises.
BLAKE3 is fast, well-designed, and gives a clean 32‑byte content ID.
Ed25519 is small, standard, and has good ecosystem support. Keys/signatures stay compact, which matters for a p2p social protocol.
Explicit typed payloads keep the wire schema readable and let me evolve each object type without tangled dynamic maps everywhere.

The goal is to keep the cryptographic and serialization layer boringly correct. Everything else can be built on top of that with confidence.

Current limits

This is still v0.1 groundwork:

No networking yet.
No storage/indexing layer.
No event log or feed processing pipeline.
Just enough error handling to make the core usable without hiding failures.

That’s deliberate. I want the “object = canonical bytes + signature + deterministic ID” story to be rock-solid before the rest of the system starts depending on it.

What’s next

The next steps are about turning the core into something people can run and poke:

Storage + indexing: persist objects, resolve by ObjectId, and build lightweight views.
Networking protocol: define how peers exchange objects and log entries (likely built on libp2p).
Tooling: a small CLI for creating/signing objects and inspecting CBOR bytes.
Interoperability tests: fixtures to prove that different implementations serialize and hash the same bytes.