Changelog

What changed in each release, and what it means at the bench.

October 8, 2026, latest release

v1.11.0

The agent bench release

Free for individuals, including at work, with no limits — every feature, Neuro and local RAG included, with no request or document cap and no license check. Companies of more than 10 employees that roll NeuroTerm out to their staff buy Commercial seats
Agent alerts and Agent resume — when Claude Code or Codex in a tab finishes or waits for you, you get a notification, the taskbar flashes and the tab is marked; after a relaunch each tab resumes its own conversation. One Agent integration setup per host installs the hooks
Agent serial access — Claude Code and Codex can read your serial tabs, wait for a boot line, list the ports and export a log; when you allow it they type into a port, reset the board or borrow the port for flashing
NeuroFlash log search — press Ctrl+Shift+F, switch to Ask, describe a symptom and press Enter; Jev finds the matching log lines through your own OpenRouter account, and secrets are redacted before anything is sent
Local shell — native PowerShell, Windows PowerShell or cmd on Windows and your own shell on Linux, as a tab next to serial, SSH and WSL
Tabs and workspaces — undo a close with Ctrl+Alt+Z, Merge into Tab, the running command on each tab, and stacked workspace cards where Ctrl+Tab jumps to the agent that is waiting
Filter and Files panels — a live list of a tab's matching lines that jumps to them; for SSH and WSL tabs a Files panel with refresh, keyboard navigation, safe delete, progress from the click and multi-file drag-out
Upgrading from 1.10 — re-index your RAG documents once (Context panel → Reindex all documents) and re-run Agent integration on each host (Settings → Coding agents)

What’s new and how to try it Release on GitHub

September 13, 2026

v1.10.0

Workspaces and session restore

Workspaces — group related sessions into named, colour-tinted workspaces, drag to reorder them, and cycle with Ctrl+Tab. A workspace you cannot see tells you when something happened in it
Your session comes back — close the app and relaunch: tabs, panes and workspaces reopen, shell sessions reconnect, each one returns to the directory it was in, and the scrollback you had is drawn back onto the screen
A collapsible tool drawer — the toolbar folds away when you want the screen, and the tab strip now flows to fit the space instead of squeezing every tab to its minimum. Quick Send reached the pop-out window too
Modern WSL rendering — NeuroTerm now ships its own up-to-date Windows console host, so WSL sessions render the way the tools inside them expect
Reconnects you can trust — each session now owns its reconnect, survives the machine waking from sleep, and a popped-out window mirrors exactly what the main window is doing
Sharper command suggestions — ghost typing now anchors to your shell prompt itself rather than guessing where the prompt ends, so suggestions land on the right column and history is recorded at the right moment
Terminal fidelity — pasting multi-line text sends one newline per line, full-screen apps can set your clipboard (OSC 52), and emoji and CJK now occupy the same width NeuroTerm and your TUI both agree on
AI that works with what you have configured — the selection menu, Analysis and Magic Input all stay available on a cloud-only setup, and reasoning models no longer leak their scratch work into results

Release on GitHub

August 2, 2026

v1.9.7

Terminal rendering, fixed at the root

The long-run garbled-text bug is gone for good — we traced the wrong-glyph corruption back to the glyph-cache refresh that was supposed to prevent it. Removing that refresh stops the corruption entirely; the renderer now recovers on its own
WSL sessions no longer leave leftover characters — running full-screen tools like Claude Code under WSL could strand stale text on screen that nothing would clear. Windows console support is now properly enabled, with an automatic repaint on scroll
New "WSL scroll redraw fix" setting — on by default under Settings › Terminal, and throttled so continuously scrolling output costs at most one cheap, flicker-free repaint per interval
Cleaner redraws in AI and TUI tools — in-place screen updates from tools like Claude Code no longer leave ghost characters behind
Split panes render reliably — panes sharing a rendering cache can no longer corrupt each other, so switching tabs or sessions keeps every pane readable

Release on GitHub

July 14, 2026

v1.9.6

Full model catalog & TUI rendering fixes

See every cloud model — the OpenRouter model list now loads the full catalog instead of falling back to a handful of hardcoded options, so you can pick from all available models
No more garbled text in busy terminals — full-screen tools like Claude Code, htop, and live log tails no longer accumulate glyph corruption during continuous output; the display now repairs itself mid-stream instead of only when idle
Background panes stay clean too — the self-healing rendering fix now works for panes running in the background, not just the one you’re looking at

Release on GitHub

July 10, 2026

v1.9.5

Long-run stability & smarter suggestions

Cleaner serial command suggestions — inline ghost-typing on serial sessions now renders as a non-destructive overlay, so suggestions never overwrite real output or leave gray residue behind
Magic Input stays out of your way — the "@" trigger no longer collides with full-screen apps like Claude Code, vim, or less, and you can now choose your own trigger character and when it activates
Fixed long-run text corruption — emoji and CJK text no longer garble after hours of use; the glyph-cache refresh was re-landed safely, without the split-pane regression it caused before
GPU re-enable now recovers every session — turning GPU acceleration back on after a rendering fault un-bans all sessions, not just the focused one
Lower CPU on busy sessions — terminal data now travels over a single streamlined path instead of being encoded and sent multiple times per byte
Steadier memory on marathon sessions — background sessions no longer stall the read loop, and per-session scrollback can shrink back down instead of only growing

Release on GitHub

June 23, 2026

v1.9.4

UI polish & rendering stability

Right-click menu stays on screen — opening the terminal context menu near the bottom edge no longer pushes the Paste option off-screen, and the Paste row now shows its Ctrl+V / ⌘V shortcut
Saved Profiles list fits any window — on short windows the profile list no longer overflows and clips the lower entries; it now scrolls within the available sidebar space
Popped-out serial sessions no longer double your typing — detaching a serial session into its own window now mirrors your echo setting, so each keystroke appears once instead of twice
Rock-solid terminal rendering — rolled back an experimental glyph-cache change that could garble or blank text in split panes when switching tabs or sessions; rendering is back to the stable, known-good path

Release on GitHub

June 14, 2026

v1.9.2

License fix & AI hardening

Fixed Pro license activation — valid license keys were being wrongly rejected as invalid due to a product-ID mismatch in the activation check. Enter your key again on this version and it activates immediately
Safer AI agent commands — the agent now blocks destructive command arguments (like deleting files or flushing routes) before they run, instead of relying on a name-only allowlist
Package-install hallucination warning — when an AI suggestion installs a package, NeuroTerm now flags it so you can double-check the name before running it
Encrypted credential storage — saved secrets are now encrypted at rest with a machine-derived key, with an atomic, corruption-resistant save path on Linux and Windows
Agent plan preview — see the agent’s intended steps before it acts, with a toggle to skip and a quick cancel
Clearer AI answer confidence — answers now show a confidence chip based on source count and quality, and low-confidence context is handled more gracefully
Reliability fixes — resolved cancel/timeout edge cases in LLM streaming and tightened auto-error handling so rapid re-submits and cancels behave correctly

Release on GitHub

June 7, 2026

v1.9.1

Per-pane input & polish

Each split pane now has its own serial input bar — type and send to exactly the session in that pane, with its own connection state, line-ending, and filter, instead of one ambiguous global input that only reached the focused pane
Filtered panes stay filtered — a split tab that left and re-entered view no longer briefly dumps the full unfiltered scrollback; each visible pane re-applies its own filter on return
Scroll lock now has a StatusBar toggle button — lock the terminal from the status bar in addition to the Scroll Lock key, with a clear active state
Spec-audit bug fixes — closing a session no longer leaks its operation context, dev builds can’t poison the “last seen version” banner, ghost-typing now honors your ranking settings, and interrupted SFTP transfers resume from the right path
USB serial reconnect wired up — disconnect banner now tracks the device fingerprint so a re-plugged adapter can reconnect
Embedding model management — pick, activate, and download embedding models from a dedicated Settings section

Release on GitHub

May 30, 2026

v1.9.0

Split-window panes

Split-window panes are now on by default — split any tab side-by-side with Ctrl+Shift+D or stacked with Ctrl+Shift+E, focus panes with Ctrl+Alt+Arrow, resize with Ctrl+Alt+Shift+Arrow, and close with Ctrl+Alt+W
Zoom a pane to fill the tab with Ctrl+Shift+Z, then restore it just as fast — great for focusing on one session without losing your layout
Broadcast input across panes — Ctrl+Alt+B fans your keystrokes out to every visible pane in a tab (tmux synchronize-panes style), with a first-time warning so you don’t fire a destructive command everywhere by accident
Drag a tab onto a pane to split or swap — drop on an edge to split in that direction, drop in the center to swap which session lives there; move sessions across tabs or pop one out into its own tab
Layout presets — jump straight to single, vertical split, horizontal split, three-pane, or four-pane grid layouts
Smarter SSH host-key handling — a Known Hosts manager with one-click re-trust when a key changes, strict verification with no silent auto-trust
Sharper terminal rendering — fixed CJK/multibyte garbling under multi-pane load, eliminated session-switch flicker on long-scroll terminals, and stopped scrollback from being truncated when switching panes
Dozens of multi-pane polish fixes — cleaner active-pane outlines, better resize handles, preserved scrollback in narrow panes, and reliable pane close without leaking sessions

Release on GitHub

May 19, 2026

v1.8.3

Tab activity V3 & right-dock polish

Tab activity V3 — residual amber pips on Claude Code TUI tabs are gone. A three-layer defense (two-window baseline confirmation, per-row periodic-redraw classifier, layout-resize invalidation) absorbs status-line ticks, post-tab-switch baseline drift, and sidebar/panel reflow while still blinking for real activity
Spinner blink preserved — the new classifier explicitly bypasses suppression for fast-cadence single-row changes, so the in-place spinner indicator still works as intended
Right-dock resize handle self-heals — dragging the right-docked Panel to a width wider than the current viewport can render no longer leaves the handle floating inside the terminal area; the stored width is clamped to the renderable max on mount and on window resize
Viewport-aware drag — keyboard and mouse resize on the right-dock now cap at the actual rendered max instead of the unconditional 600 px ceiling, so drag-delta matches visual-delta from the first pixel
Right-docked feedback banner re-centers — the banner now centers over the main column instead of the full viewport when the sidebar or right-dock is expanded
Right-dock terminal fits cleanly — xterm characters no longer clip at the right edge during drag-resize; fit() now fires on every width tick (gated to the active session) and coalesces SIGWINCH so non-TUI shell scrollback is protected
Right-dock overlays cleared — idle toast, AI toast, update notification, and the @-trigger Magic Input palette now slide clear of a right-docked panel instead of being hidden underneath

Release on GitHub

May 17, 2026

v1.8.2

Pop-out panel & tab activity fixes

Detached SessionWindow panel layout fix — in popped-out windows the bottom Panel no longer covers the entire terminal, and the resize-drag handle is reachable again with the separator restored between terminal and panel
Tab activity false-positive suppressed — synchronized-output TUIs like Claude Code and OpenCode no longer trip the inactive-tab amber dot on transient mid-parse buffer mutations; the dot now only fires after the parse stream settles

Release on GitHub

May 16, 2026

v1.8.1

Drag-and-drop panel docking

Drag-and-drop panel docking — grab the Panel header grip and drop it on the right edge to dock vertically (or back to the bottom). Agent chat state, scroll positions, and in-flight LLM streams all survive the dock change without reload
Persistent dock layout — the chosen dock side and width are remembered across restarts, with a single StatusBar toggle that operates on whichever dock is active
Polished drop zones — dock-direction icons, single-line labels, wider hit areas, and clearer over-state feedback make the drop target obvious before you release
Terminal refit fix — the terminal no longer gets stuck treating a collapsed panel as expanded, eliminating one source of resize-loss when toggling the bottom panel
StatusBar Blocks icon cleanup — hides the numeric count next to the icon (still available in the hover tooltip; Alt+5 unchanged)

Release on GitHub

May 15, 2026

v1.8.0

Command Blocks + tab activity v2

Command Blocks panel — capture, browse, and replay shell commands in a dedicated panel with one-click navigation back to where they ran in the terminal. Status bar button + Alt+5 hotkey for quick access
Boot-stage virtual blocks — pattern markers like "U-Boot Start" or "Systemd Ready" appear in the Blocks panel as virtual rows, even without OSC 133 shell integration
Newest-first ordering, inline expand, one-click bookmark — newest commands sit at the top; double-click a row to expand its output inline (height fits content, no fixed cap); single-click any row’s bookmark icon to toggle
Context menus everywhere on a block — right-click any colored gutter strip in the terminal or any row in the panel to open the same block actions menu (Copy command, Copy output as Markdown, Send to AI, Rerun, Bookmark, Delete)
Absolute timestamps + tighter rows — every block row shows the exact HH:MM:SS it finished (relative "5m ago" available as a hover tooltip), with tighter row height for higher information density
Tab activity dot v2 — tabs blink while content streams and use viewport content-hashing to distinguish real output from cursor blinks, so the dot only lights up when something actually changed
Memory leak mitigation — long-running sessions hold tighter on memory

Release on GitHub

April 13, 2026

v1.7.3

What’s New polish & marker stability

Marker stability — pattern markers now survive terminal resize and TUI redraws instead of disappearing when xterm.js disposes IMarkers
What’s New history — re-opening the panel shows the last 5 releases instead of only the current version
Version button always clickable — the StatusBar version chip re-opens What’s New after dismiss, with a polished filled-background style
What’s New visual polish — improved version glow, section title hierarchy, and list marker styling for better readability

Release on GitHub

April 12, 2026

v1.7.2

Panel resize & TUI stability

Panel toggle no longer shrinks terminal rows — fixed multiple resize race conditions that caused line loss when opening or closing the bottom panel
TUI app stability — sync-output apps like htop and vim now maintain correct dimensions during panel and sidebar transitions
What's New panel — post-update feature discovery via a StatusBar chip and slide-out panel so you never miss new features
Error explanation improvements — auto-dismiss stale cards, archive and restore dismissed explanations, scroll-to-error from history
WSL cursor offset fix — resize-on-connect and updated estimation constants eliminate cursor misalignment on Windows Subsystem for Linux
LLM crash prevention — serialized native LLM access prevents exit-code-3 crashes when multiple error explanations trigger simultaneously
Minimap rendering fix — cleared shadow-blur residue that left artifacts at the bottom edge of the minimap canvas
Free-tier AI enforcement — Neuro Agent sessions now correctly respect free-tier usage limits

Release on GitHub

April 10, 2026

v1.7.1

Gemma 4 & error annotations

Gemma 4 model support — upgraded llama-cpp-2 to 0.1.143 with native Gemma 4, Llama 4, and 134+ architecture support
Structured error annotations — inline error explanations with What/Cause/Fix sections and red line highlighting directly in the terminal
Error annotations in Neuro tab — auto-detected errors are saved and browsable in the AI panel with per-session toggle
Jinja template fallback — models with unrecognized chat templates now render correctly via Jinja/minja instead of failing
Speculative decoding fix — resolved ISWA decode crash affecting Gemma 4 and other interleaved sliding-window models
Agent panel polish — fixed hover states, button visibility, idle content wipe, and adaptive CPU timeout for slow prompt processing
Key input lag fix — deferred highlighting to RAF flush and optimized ghost typing retries under high throughput

Release on GitHub

April 5, 2026

v1.7.0

Neuro Agent & AI diagnostics

Neuro Agent — an always-on AI pilot that monitors your serial session in real time, detects errors, and explains them using RAG-augmented context
Five new AI-native features: real-time error explanation, device knowledge auto-accumulation, datasheet register/error code decode, autonomous device diagnostics, and intelligent context assembly
Multi-turn agent conversations — ask follow-up questions to the Neuro Agent with full conversation history
Neuro/Agent mode toggle — switch between AI chat and autonomous agent mode directly from Magic Input
Status bar agent indicator — see at a glance whether the Neuro Agent is active and monitoring
TUI rendering fixes — resolved stale characters, glyph corruption, and refresh timing issues in full-screen terminal apps
SSH reconnection fix — passphrase-protected keys with Remember enabled now reconnect reliably
Performance boost — eliminated Zustand store cloning on high-frequency session updates

Release on GitHub

March 28, 2026

v1.6.2

SFTP tooltip upgrade

Custom hover tooltips for SFTP and WSL file browsers — replaces unreliable native title attributes with rich, positioned tooltips showing filename, path, size, permissions, modified date, and symlink target

Release on GitHub

March 23, 2026

v1.6.1

Smarter idle & SFTP polish

Enriched idle summaries — structured metadata with event timeline, error clustering, and boot detection replaces raw terminal dumps
Platform-aware idle analysis — system prompts tuned for ESP-IDF, Zephyr, Linux, and network targets
SFTP file tooltips — hover any file to see full name, size, permissions, modified date, and symlink target
Terminal clear now properly resets filtered viewports — fixes stale data persisting after clear in show-only filter mode

Release on GitHub

March 19, 2026

v1.6.0

Cloud AI & crash resilience

OpenRouter cloud AI backend — connect to 300+ cloud models (GPT, Claude, Gemini, Qwen) alongside the local llama engine
Multi-turn AI chat — Magic Input now remembers conversation context within each session for follow-up questions
Crash reporting — panic hooks capture backtraces to a persistent crash log with automatic 1MB rotation
Smarter RAG retrieval — multi-query decomposition splits complex questions, timing-parameter boost for datasheet lookups
Hardware synonym expansion — RAG searches automatically match MCU terms like baud/BRR, clock/RCC, interrupt/IRQ
Duplicate session preserves directory — new SSH/WSL sessions open at the source session’s working directory
Quick Send editor polish — redesigned 3-row inline editor with proper layout and wider command input

Release on GitHub

March 9, 2026

v1.5.0

Multi-select transfers & quick macros

Multi-select file transfers — Ctrl+click and Shift+click to select multiple files in SFTP and WSL browsers for batch download or upload
Transfer progress stats — live speed (MB/s), ETA, and file size displayed during SFTP and WSL file transfers
Quick Send macros — assign F1–F4 hotkeys to frequently used commands, with built-in defaults per job mode (Linux, MCU, Network)
Quick Send toolbar button — compact dropdown replaces the previous strip, consuming zero terminal space
SFTP reconnect after hibernation — single Reconnect button handles all disconnected sessions at once
Upload Files context menu — upload files to WSL targets directly from the browser right-click menu
Background context menu — right-click empty areas in SFTP/WSL browsers for directory-level actions

Release on GitHub

March 8, 2026

v1.4.0

RAG reliability & filter stability

Smart RAG retrieval pipeline — multi-stage search with FTS fallback eliminates silent failures on SVD and hierarchical documents
CMSIS-SVD file import — import peripheral register maps for AI-assisted embedded debugging
KV cache quantization and speculative decoding — new LLM performance settings for faster inference
Matryoshka embeddings — configurable embedding dimensions for flexible RAG accuracy vs speed trade-offs
Snapshot-based filter restoration — toggling filters no longer loses ANSI colors or inflates line counts
Per-session resize controller — fixes SIGWINCH storms that caused shell redraws during layout changes
Pop-out window reliability — filter state preserved across pop-out/reattach, stale subscriptions cleaned up
Windows >2GB GGUF support — vendored llama-cpp-sys-2 patch fixes 32-bit file offset overflow
Qwen3.5 model support with native chat template formatting

Release on GitHub

March 2, 2026

v1.3.5

Credential store & smart tabs

SSH password and passphrase persistence — credentials saved securely with OS keychain integration
Smart tab overflow — tabs gradually shrink as sessions increase, with scroll arrows when needed
Dense tab mode activates automatically when space is tight, keeping the active tab readable
SFTP reconnect after hibernation — wake recovery shows Reconnect button instead of blank state
WSL features hidden on non-Windows platforms for a cleaner interface
Redesigned New Session button with contextual styling based on session count

Release on GitHub

February 27, 2026

v1.3.4

Pill tabs & Linux packaging

Redesigned session tabs — new pill-shaped design with color accent stripes and discrete status indicators
Linux release builds — AppImage, .deb, and .rpm packages with Vulkan GPU support
Scrollbar visibility restored after xterm.js v6 upgrade
Pop-out windows now auto-switch to the nearest tab instead of leaving a dead tab active
Show-only filter mode displays a split input bar and properly clears the viewport
SSH encrypted key connections no longer leave orphaned sessions
Memoized mood ring selector prevents unnecessary tab re-renders

Release on GitHub

February 24, 2026

v1.3.3

Tab reorder & SSH rendering fixes

Reliable drag-and-drop tab reordering — works correctly with pop-out windows and overlapping tabs
Full-screen SSH apps (vim, htop, less) now render correctly with proper PTY terminal modes
Improved split input bar visibility when unfocused with accent strip and stronger borders

Release on GitHub

February 24, 2026

v1.3.2

Connect flow & popout fixes

Connect button on all new session modals — start sessions directly without saving a profile first
Embedding model now auto-downloads and configures alongside LLM tier selection
Pop-out windows render ANSI colors correctly in release builds via WebGL
Profile rollback on connection failure prevents retry dead ends

Release on GitHub

February 22, 2026

v1.3.0

Model browser & tab awareness

HuggingFace model browser — search, preview, and download GGUF models directly from Settings
Mood ring for session tabs — color-coded glow shows error/warning activity at a glance
Duplicate Session from tab context menu — quickly clone any connection
HEX mode badge in status bar with pulsing indicator
Simplified embedding model setup — Nomic Embed auto-bundles with LLM downloads
Fixed false activity indicators on inactive tabs caused by split ANSI sequences
Free tier RAG document limit now enforced on all import paths
GPU detection fix for Vulkan builds on NVIDIA hardware

Release on GitHub

February 19, 2026

v1.2.0

Smarter AI tools & pattern detection

Per-invocation context — add custom notes to any Neuro Tool run without changing the tool definition
RAG document @-mention in Neuro Input — reference specific docs when issuing AI commands
Panel markers now send surrounding terminal context to Neuro Tools for better analysis
SFTP sidebar reconnect button and quick-connect buttons when no session is active
Expanded error and warning patterns — now catches critical, fatal, segfault, OOM, corruption, timeout, deprecated, and more
Improved Job Mode markers with Systemd Ready for Linux and unified MCU crash/fault coverage
Custom Neuro Tools now use a neutral prompt — no more forced terminal diagnostics framing
Pattern case sensitivity preserved correctly when editing existing regex patterns

Release on GitHub

February 18, 2026

v1.1.0

Configuration & reliability

Full configuration export/import — settings, profiles, AI tools, RAG, and patterns in one file
Pattern overrides (mode-specific markers and problems) now included in config exports
Backward-compatible v1 config migration to new v2 format
Recording button decoupled from auto-log setting — no more accidental setting changes
Fixed empty log files on second recording session
Fixed log timestamps appearing when timestamp mode is off
Delta timestamp mode for log files using real-time intervals
Settings panel simplified — no more hidden advanced sections
Credential hardening — SSH passwords and passphrases no longer saved to disk

Release on GitHub

February 17, 2026