v0.27.2

[0.27.2] - 2026-10-27

Upgrading from 0.27.1. No migration and no change to your cards, sources or settings. This
release fixes the Neural Interface on every install: the card's Allow / Use the local model taps,
date answers and Refine work again. Nothing about research, consent scope or what leaves the device
changes, so there is no new held-out safety run.

Fixed

• Neural Interface: answering a card's question (Allow a cloud model, Use the local model, a date) and
  Refine failed with "That didn't go through" since 0.26.0 (#510): the web app sent the Content-Type
  header twice and the server refused the body. The request helper now owns that header.
• Neural Interface: a build whose step is set to Local (or Cloud) with no model routed on that side
  said "The request couldn't be understood well enough to build from". The card now says that no
  model is routed for building cards and where to fix it (Settings > Models).
• Global Library: a library URL that answers HTTP 404 (nothing published there) is reported as such,
  and any other HTTP status is named ("the library host answered HTTP 403"), instead of "couldn't
  reach the library host".

What's Changed

• build(packaging): bump manifests to v0.27.1 by @SecureCloudGroup in #560
• NI: one Content-Type per request, an honest sentence for a missing build model, a library 404 named as such by @SecureCloudGroup in #561
• Release 0.27.2: changelog section for the Allow-tap fix by @SecureCloudGroup in #562

Full Changelog: v0.27.1...v0.27.2

Renderer sidecar image: ghcr.io/securecloudgroup/smartbrain_3000-render:0.27.2@sha256:805957e626cd6c77dc2bf140dab6fcdb6c1b2e0698f83ac72a64228990c2bd26