Skip to content

Claude Assist 2.5.3

Choose a tag to compare

View all tags
@kvrnl-admin kvrnl-admin released this 02 Jul 00:12
· 4 commits to main since this release
v2.5.3
dffac63

Claude Assist 2.5.3

Hotfix: fixed a bug introduced in 2.5.2 that made Neural Link reply with an error/garbled text to every message. Chat works normally again.

Previous release notes:

Neural Link upgrades: it can now look things up on kvrnl.io (latest version, changelog, docs, account help), file a bug report to the KVRNL team for you (with an ATTACH button to include a screenshot), and there's a new AUTO-SCROLL toggle in its header to turn chat auto-scrolling on or off.

Previous release notes:

Tidied up the Neural Link header bar — removed the lightning-bolt icon, the memory counter, and the model badge for a cleaner look.

Previous release notes:

Neural Link is now more powerful: it can read and search your files and projects (read-only) to answer questions, explain code, and help you find things — running on your own Claude Code CLI. It stays strictly read-only: it can never write, edit, delete, run commands, or go online. For actual changes, start a real Claude Code session.

Previous release notes:

Neural Link's memory is now hardened to survive restarts, crashes, and app updates: it lives in a dedicated folder outside the install location (so updates never touch it), saves are crash-safe, and it keeps a backup copy that auto-restores the memory if the main file is ever corrupted.

Previous release notes:

Fixed a bug where Neural Link could mistake the maker's name (KVRNL) for the user's own identity and greet people as KVRNL. Neural Link now never treats the app's maker as the user, and any profile that was already affected is cleaned up automatically on next launch.

Previous release notes:

Clicking a sidebar section now auto-collapses the Neural Link panel to its header bar, so switching tabs tucks the assistant out of the way (one click on its header reopens it).

Previous release notes:

Neural Link is now a proper in-app assistant: it knows how to explain every feature of Claude Assist and walk you through using it, remembers who you are between chats, and keeps a small optional profile (name, business, email, role, what you're working on) — it'll ask for those gradually and use them to personalize help. All optional, never pushy.

Previous release notes:

Neural Link now has its own private home folder (C:\ClaudeAssist\NeuralLink). Its chat sessions run there — fully isolated from your real project folders — and its memory lives there so it stays separate and secluded while still remembering you.

  • The Agent Hub has been fully removed from Claude Assist (it's now its own separate app).
  • Removed the leftover Codex CLI integration and the Codex line on the About screen; the app no longer looks for or connects to Codex at all.

Previous release notes:

Neural Link runs exclusively on Claude Sonnet 5 — one fast, capable model, no picker.

Neural Link privacy + memory fixes:

  • Neural Link no longer opens inside your most recent project folder. It had been auto-attaching to your last-used project (with read access to its files) — it now has NO access to your files or folders at all and is purely a chat + session-recall assistant. If you want it to read or change code, start a real Claude Code session.
  • Neural Link now actually remembers who you are — it learns and saves your name/details, and the memory counter updates as it learns.

Also includes the 2.3.1 update-button hotfix and the full 2.3.0 bug-sweep release below.

  • Fixed a licensing check that could lock everyone out if our server had an outage (the app now correctly falls back to the 14-day offline grace).
  • Voice notifications (read aloud) now actually work.
  • Backup fixes: retention cleanup could delete backups belonging to a similarly named folder; failed backups no longer leave half-written archives; deleting a backup now asks for confirmation.
  • Quitting the app no longer leaves a hidden background process running.
  • Popup answers can no longer get mixed up between multiple Claude Code sessions.
  • The Claude Code integration no longer crashes on emoji or special characters.
  • Popups now position correctly on multi-monitor setups.
  • Your notes, tasks, timeline and settings files are now crash-proof (atomic saves) and survive corruption instead of being silently wiped.
  • Crash/support reporting now routes through the KVRNL backend instead of a hardcoded endpoint (more secure, rotatable server-side).
  • Removed leftover Discord Rich Presence code from an older version.
  • Dozens of smaller crash fixes and edge-case guards throughout.
Assets 3
Loading