BioRouter v1.86.0

Biorouter v1.86.0 Release Notes

Release Date: June 2026
Repository: github.com/BaranziniLab/biorouter

This is a feature-and-fix release. It adds one-click "Restart & Update" auto-update for macOS, multi-monitor screen capture with vision input, agent-readable UI-automation errors, and new built-in extension/skill authoring skills. It also adds a Capabilities settings section for foundational built-ins and hardens the Agent Drafter preview. No breaking changes — existing installs upgrade in place.

Downloads

Platform	File	Install
macOS (Apple Silicon)	BioRouter-1.86.0-arm64.dmg	Open the DMG and drag BioRouter.app to /Applications
macOS (Intel)	BioRouter-1.86.0-x64.dmg	Open the DMG and drag BioRouter.app to /Applications
Windows (x64)	BioRouter-win32-x64-1.86.0.zip	Unzip and run BioRouter.exe
Linux Ubuntu / Pop!_OS (x64)	biorouter_1.86.0_amd64.deb	sudo dpkg -i biorouter_1.86.0_amd64.deb
Linux Fedora / RHEL (x64)	BioRouter-1.86.0-1.x86_64.rpm	sudo rpm -i BioRouter-1.86.0-1.x86_64.rpm
Linux — CLI only Ubuntu / Debian (x64)	biorouter-cli_1.86.0_amd64.deb	sudo apt install ./biorouter-cli_1.86.0_amd64.deb
Linux — CLI only Fedora / RHEL (x64)	biorouter-cli-1.86.0-1.x86_64.rpm	sudo dnf install ./biorouter-cli-1.86.0-1.x86_64.rpm

The two CLI-only packages install just the headless binaries (biorouter + biorouterd) with no Electron/GUI payload — ideal for servers, HPC nodes, and containers. (Local models need glibc ≥ 2.35 — Debian 12+/Ubuntu 22.04+.)

What's New

One-click "Restart & Update" (macOS)

Updating no longer means opening the download page, grabbing the DMG, and dragging to Applications. A single button now downloads the new release in the background and, on click, restarts Biorouter on the new version in place. Your settings, sessions, and extensions are preserved — only the app bundle is swapped; ~/.config/biorouter is untouched. Both the update modal and the Settings update section drive the same one-click flow with a live progress bar. The release pipeline now also ships the signed zips and a latest-mac.yml manifest that the updater needs.

Multi-monitor capture + vision input

The Computer Controller and Developer screen-capture tools now enumerate connected displays (index, name, resolution, position, scale, primary) on macOS, Linux, and Windows, so a multi-monitor setup can re-capture any screen by index, and list_windows accepts a window-title substring. Xiaomi MiMo (v2.5 / v2) is now declared image-capable, so it can consume screenshots like the Anthropic and OpenAI vision providers.

Extension / skill authoring skills

develop-biorouter-extension and develop-biorouter-skill are now real built-in SKILL.md skills for authoring Biorouter extensions and skills.

Capabilities — on by default, managed in one place

The foundational built-ins — Developer, Extension Manager, Skills, Todo, Memory, Knowledge — are now first-class Capabilities in Settings, enabled by default:

• Default-enabled on upgrade. If you upgraded from an older build where some of these were left off, this version turns them back on for you (once) so the agent has its core abilities out of the box. You can still disable any of them in Settings → Capabilities, and that choice sticks.
• Out of the chat extension list. Capabilities no longer clutter the per-conversation extension dropdown in the chat box and the dashboard — that list now shows only the optional built-ins and the extensions you install yourself. Capabilities are toggled only in Settings, where they belong.

Agent reliability

Backend agent-loop improvements (shared by the CLI and the desktop app), found while stress-testing the agent across 100 real multi-file projects:

• Continue-on-truncation. When a provider cuts a response off at its output-length limit (finish_reason="length") with no tool call, the agent auto-continues the turn instead of stopping on a half-written response — gated strictly on the explicit length signal and bounded so it can't loop.
• No more "keep going" loops when stuck. An earlier mid-task "keep working until your todos are done" nudge could repeat itself when the agent was genuinely blocked (e.g. an unrecoverable provider error) without resolving the cause. That hard-coded nudge is removed; "don't stop while work is unfinished" is now handled only by the bounded, user-configurable Stop-hook and /goal mechanisms (which give up when progress stalls).
• The turn-ending finish_reason is surfaced in the logs, so "finished" vs "cut off" is distinguishable.

CLI quality-of-life (interactive TUI)

• Scroll the whole conversation. The mouse wheel now scrolls the conversation history instead of the terminal's own buffer (which used to expose the pre-launch banner). Native text selection still works with your terminal's bypass modifier (Option on macOS Terminal/iTerm2, Shift elsewhere); PageUp/PageDown scroll too.
• Tidier tool output. As tool calls pile up, older tool results collapse to a one-line summary so only the current result stays expanded — the conversation no longer gets buried under walls of old output.
• See your queued messages. Messages you type while the agent is working now appear in full in a "queued" pane (not just a count), so you always see what runs next, in order.

MiMo vision fix

Only the multimodal mimo-v2-omni model is now advertised as image-capable; the text MiMo models (which reject image input) no longer get fed screenshots — and, defensively, image content is stripped from requests to a non-vision MiMo model. This fixes a stuck loop where a screenshot/vision step would 404 (No endpoints found that support image input) and strand the agent.

What's Fixed

• UI automation: failing osascript / PowerShell scripts now surface as real tool errors with stderr and exit status (instead of a fake "success" that made the model retry blindly), and are guarded against long hangs.
• Code execution: binary/blob resources (e.g. Auto Visualiser ui:// HTML blobs) are collected out-of-band and appended for inline UI rendering, with no duplicate payload for text resources.
• Agent Drafter: preview auto-resize, in-app chat protocol, and sandbox-safe send.

Upgrading

No action required beyond installing the new build for your platform from the table above. Settings, sessions, knowledge bases, and configured providers are preserved. On macOS, in-app "Restart & Update" handles the swap for you.

Changes Since v1.85.4

• feat: multi-monitor capture + vision input, agent-readable UI-automation errors, dev skills
• feat(updater): one-click "Restart & Update" auto-update for macOS
• feat(settings): add Capabilities section for foundational built-ins
• fix(agent-drafter): preview auto-resize, in-app chat protocol, sandbox-safe send
• feat(agent): continue-on-truncation when a turn is cut off at the output-length limit
• feat(agent): log finish_reason at turn-end (done-vs-stopped observability)
• feat(capabilities): default-enable foundational capabilities on upgrade; hide them from the chat extension dropdown (managed in Settings)
• fix(agent): remove the hard-coded "unchecked todos" continuation that could loop when the agent was stuck; defer to the bounded Stop-hook / /goal mechanisms
• fix(providers/mimo): advertise vision only for mimo-v2-omni; strip image content from requests to text-only MiMo models (fixes the image-404 stuck loop)
• feat(cli/tui): mouse-wheel scrolls the conversation; collapse older tool outputs; show queued messages in full