What 1.0 actually means
The eos-platform meta-distribution has been usable for nearly a year, but every release leading up to today carried at least one breaking change in the manifest schema, the package server URL, or the build-system overlay. Today's 1.0 freezes all three. The on-disk manifest format, the registry protocol, and the eos-platform-build CLI surface are now under semantic-versioning guarantees: minor releases will not break working profiles.
Internally, that meant retiring the legacy profile.toml dialect (last seen in 0.4) and converging on the profile.yaml schema first published in 0.7. The migration tool ships in this release and runs idempotently.
One toolchain, fourteen components
eos-platform's defining promise is that you don't pick and choose your way through fourteen separate Cargo / CMake / Bazel projects. You declare a profile — say, "smart-home gateway with onboard inference" — and a single build invocation produces signed, attested, reproducible images for every target board you've registered:
$ eos-platform build profile.yaml --target=stm32h7-eval --target=rk3588-soc
[+] resolving 14 components from registry.embeddedos.org
[+] EoS RTOS 4.2.0 (cached)
[+] EAI 0.9.0 + INT4 runtime
[+] eBootloader 1.4.0 (signed)
[+] reproducible build: hash deadbeef... matches upstreamThe --reproducible flag is on by default and asserts byte-identical output against the foundation's continuously-rebuilt reference images.
Stable APIs, not stable everything
What we've stabilized: the manifest schema, the CLI, the public C ABI between EoS RTOS and EAI, the JSON-RPC API exposed by EoStudio. What we have not stabilized: the wire protocol between EoSim and external HIL bridges (still 0.x), and the eDB on-disk page format (the encryption-at-rest work — see our deep-dive — bumps it to v3 next quarter).
What's next
Two large rocks for the next six months: a hosted package mirror in Europe, and an opt-in telemetry channel that gives the foundation an aggregate view of which profiles are actually deployed. As always, both are governed by the foundation charter.
Read next
EAI 0.9 ships INT4 LLM runtime — 11 tok/s on a Cortex-M85
EAI's new quantized inference path squeezes a 1.3B-parameter model into 312 MB of flash and runs at interactive speed on a 480 MHz microcontroller. We dig into the kernel scheduler that made it possible.
ENI's 1,024-channel pipeline: deterministic spike sorting in 800 µs
How the Embedded Neural Interface stack moves a thousand-electrode array through filtering, sorting, and decoding inside a single RTOS frame — and why the hardest part wasn't the math.
eBootloader secure boot: a measured-launch walkthrough
An end-to-end tour of eBoot's chain of trust — root-of-trust keys, immutable stage 0, signed manifests, anti-rollback counters, and the runtime attestation hooks EAI consumes during model load.