Introduction

e-serde is part of the Eager (e-) stack by damvolkov — a family of local-first, production-grade components. e-serde is its serialization layer.

Why

Python has never lacked config loaders; it lacks a fast, type-safe, uniform one. The usual options are either a pure-Python parser that is easy but slow, or a fast C/Rust codec that is quick but format-specific and typing-optional. Choosing formats means choosing five different APIs and five different failure modes.

e-serde's thesis: take the speed and robustness of C and Rust — the same engines that already power the fastest JSON and TOML libraries — and put them behind one small surface, so a config file becomes a validated Python object with a single call.

Built with open-source models

The codecs are not reimplemented here; they are the state of the art, wrapped:

Concern	Engine	Language
JSON decode	msgspec	C
YAML	saphyr (YAML 1.2 core schema)	Rust
TOML	toml (toml-rs org)	Rust
JSONC	jsonc-parser (from Deno)	Rust
INI	rust-ini	Rust
CSV · TSV	csv (BurntSushi)	Rust
Validation	msgspec.convert (Structs/dataclasses)	C

Everything Rust lives in a single PyO3 extension module, compiled by maturin, released as prebuilt wheels. The only runtime dependency is msgspec.

What it is not

• Not a config framework — no env vars, no layering, no schema registry. It decodes bytes into objects. Reach for that on top if you need it.
• Not losslessly round-trippable for every format by design: JSONC comments are dropped on dump, TOML has no null, YAML rejects non-scalar keys. These limits are explicit, not surprising.

Next: Installation →