Skip to content

v0.27.1

Choose a tag to compare

@github-actions github-actions released this 27 Oct 01:25
· 18 commits to main since this release

[0.27.1] - 2026-10-26

Upgrading from 0.27.0. No migration and no change to your cards, sources or settings. This
release changes what a paired phone (or a browser on a LAN address) may do:

  • A phone can now ask for a Neural Interface card and work the cards it has, and may tap the four
    fixed-choice consents (allow a cloud model on a card, Activate a card's other sites, Keep looking,
    the repair policy). Each tap is checked by the Desktop and shown in Activity as "from paired
    phone". A cloud-model allowance or a Keep looking from a phone draws one unit of the daily
    re-research ceiling (SMARTBRAIN_NI_RERESEARCH_DAILY, default 10) and is refused for the day once
    it is spent.
  • Settings writers that answered a phone before this release no longer do: provider keys, the
    passphrase change, local-model endpoints, model routing and Neural Interface step models, the
    web-search provider and the self-improvement switches now refuse anything but the Desktop. The one
    phone-visible surface was the chat page's "connect the detected local model" button. If you relied
    on any of those from a phone, do them on the Mac from now on.
  • Every Desktop-only refusal reads "Finish this on your Mac: it can't be done from a paired phone or
    a LAN browser." (A phone consent refused because the day's ceiling is spent says so instead.)
  • SECURITY.md now states the widened reach of the signaling node: whoever controls that node's
    code can do what a phone can do (including charging cloud-model allowances to your key, within the
    daily ceiling), and nothing more. Running your own node via SMARTBRAIN_SIGNALING_URL moves that
    trust to you.

Neural Interface research itself is unchanged, so the held-out safety numbers from 0.27.0 stand.

Added

  • A paired phone can drive Neural Interface cards. It can ask for a card (the request text
    only) and work a card it already has: Retry, Cancel, Fix, Refine, Reopen, Stop looking,
    filling a non-secret value, the brief-summary switch, and the findings and measure views. It
    can also tap the four fixed-choice consents: allow a cloud model on a card, Activate a card's
    other sites, Keep looking, and the repair policy. Pasted links, typed answers, keys, trust
    changes, feeds, MCP settings, backups, pairing and updates stay on the Desktop; a phone that
    tries one is told "Finish this on your Mac" instead of a bare refusal. Every action a phone
    takes is recorded in Activity with a "from paired phone" note (a browser on a LAN address
    reads "from LAN browser"); the audit API gains a via field. A cloud-model allowance or a
    Keep looking started from a phone draws one unit of the daily re-research ceiling
    (SMARTBRAIN_NI_RERESEARCH_DAILY), and is refused for the day once it is spent. See
    Remote access and the trust-model note in
    SECURITY.md: the phone app is served by the signaling node, so whoever controls that node's
    code can now do what a phone can do, and nothing more.

Security

  • Settings writers now refuse a paired phone or a LAN browser (they never should have
    answered one): provider keys, the passphrase change, the local-model endpoints (Ollama, MLX,
    Claude Code, voice), model routing and Neural Interface step models, the web-search provider,
    and the self-improvement switches. Settings is a Desktop-only page already; the one
    phone-visible caller was the chat page's "connect the detected local model" button, which now
    reads "Finish this on your Mac".

Changed

  • Internal: Desktop-only is now declared on the route (dependencies=[DESKTOP_ONLY]) instead of
    called inside it, and a test partitions every API route into Desktop-only, partial and
    phone-ok, so a new route picks its authority class deliberately. The refusal fires before
    input validation.

What's Changed

Full Changelog: v0.27.0...v0.27.1

Renderer sidecar image: ghcr.io/securecloudgroup/smartbrain_3000-render:0.27.1@sha256:016060ad3b7a6070ad1cc91a2c81ca3a073db0a7c736aef6d4c3341598ddf7ba