super.engineering

Detailed report

super.engineering coordinates coding agents on local repositories and your own remote hosts. It suits a workflow centered on a Mac. SSH execution is documented, but reliable handoff between laptops is not established.

Research checked November 13, 2026. Most official documentation is undated, so the access date does not establish when a feature shipped.

Product and workflow

The product is super.engineering, made by Super Dot Engineering LLC, with the official GitHub identity superdoteng. Its founder previously called it Superconductor, linking the same domain in a LinkedIn launch post. It is separate from Conductor at conductor.build.

It provides one workspace to coordinate and review your existing agents, using your subscriptions, repositories and compute. The Rust interface provides terminals, rendered chat, task-specific Git worktrees, configurable layouts, and picture-in-picture windows. A worktree is a separate checkout that lets one task change files without disturbing another task’s checkout. The vendor advertises sub-0.5-second startup and 160 MB idle memory. These are vendor measurements, not independently verified benchmarks. Official overview

Agents, models, and payment routes

The detailed provider table is more useful than the homepage’s “15 providers” count. Stable Chat UI and terminal integrations include Claude Code, Codex, Cursor, OpenCode, Pi, Oh My Pi, and Grok. Experimental integrations include Kimi Code, Prime Agent, Vercel fx, jcode, and Antigravity. Copilot, Factory Droid, Kiro, Qwen Code, Hermes, and legacy Gemini are listed as experimental terminal integrations.

You install and authenticate the underlying command-line tools yourself. Existing subscription authentication or your own API keys work where that provider permits them. super.engineering does not supply universal model access or proxy requests. Model choices and reasoning controls depend on the provider and adapter. Custom CLI agents are also supported. Provider documentation

Your laptops, devices, and headless servers

The desktop download supports macOS 14+, including Apple Silicon and Intel. Linux desktop builds support x86_64 and ARM64, require glibc 2.35+, Wayland or X11, and Vulkan. Linux lacks the embedded browser and browser automation available on macOS. Windows is planned, without a published date. Mac download, Linux requirements

For your headless servers, the important feature is experimental SSH workspaces:

• Your Mac displays the interface. Git, worktrees, terminals, and supported agents execute on your Linux or Mac host.
• Ubuntu is the documented Linux baseline. Both processor architectures are supported.
• Provider tools must be installed and authenticated remotely. SSH key authentication and forwarding are needed for full integration.
• A managed sc helper supplies live status and hooks.
• Structured PR actions normally use credentials on your Mac. Agent-driven PR commands run remotely.

This is your own remote compute, not a vendor-provisioned cloud sandbox. SSH connections originating from Linux desktops remain unverified in the documentation. SSH architecture

Session recovery limits

The app restores workspace layouts and binds chats to provider resume identifiers where available. It can fall back to a fresh session if provider history cannot be validated. That documents conversation recovery, not guaranteed survival of an active process through sleep, network loss, or app termination. I found no documented second-laptop takeover or synchronized desktop session store. Session restore

The TestFlight mobile app supports iPhone and iPad layouts. It can continue chats, answer approvals, open terminals, review changes, and commit or push. The Mac must remain awake and running the app. Dedicated PR creation and merge controls are absent. Connections use the local network or Tailscale. Other private networks need manual configuration. Direct phone-to-headless-Linux control is not established. Mobile capabilities, Network routes

Workflow and orchestration

The orchestration is more than arranging terminal panes. A lead agent can create visible sessions, route work across providers, collect results, and coordinate implementation and review through sc. The documentation distinguishes explicitly requested app orchestration from a provider’s native subagents. Enabling orchestration does not automatically authorize every task to fan out. Orchestration contract

Review stays attached to its worktree, including diffs, line comments, agent replies, checks, and the next Git action. Findings can go to the current chat, another chat, or background work. GitHub, GitHub Enterprise, GitLab, and Bitbucket Cloud have forge integrations. Tests run through agents, terminals, repository scripts, or CI. There is no evidence that orchestration itself guarantees correct testing. Worktrees also separate files, not operating-system permissions or shared databases. Review workflow, Forge support

Setup, security, license, and price

Setup involves installing the desktop app, authenticating agent CLIs, opening repositories, and optionally configuring forge CLIs and SSH hosts. No desktop account is required.

The vendor says it does not collect source code, prompts, conversations, outputs, or provider keys. Anonymous product telemetry is optional. Update checks occur hourly and include daily aggregate usage markers independently of the telemetry setting. Model providers still receive whatever their agents send. These are architectural and policy claims, not an independent security audit. Privacy policy

The app is proprietary, with a revocable license for personal or internal business use. Running it locally and using your own servers is supported. An open-source, self-hostable central orchestration service is not offered. The public GitHub repository is a feedback tracker, not application source. Terms, Repository

Published software price: $0

The terms say sponsorship is voluntary and unlocks no features, higher limits, or priority support. Agent subscriptions, API usage, and your infrastructure remain separate expenses. Enterprise rollout and support are quoted by team size and needs, with no public seat price. The enterprise offer and “sponsorship is the only charge” wording are not fully reconciled publicly. Pricing terms, Enterprise scope

Development and user reports

Current downloads are nightly builds. Public issues and maintainer replies demonstrate activity, but do not establish production reliability or adoption scale.

Positive evidence is mostly vendor-curated testimonials praising speed, interface quality, and replacing terminal-heavy workflows. I could not independently retrieve the linked X posts to verify dates and context. Treat these as selected endorsements, not consensus. Testimonials

Concrete negative reports include:

• November 4, 2026: a user reported freezes after closing the lid or docking. Maintainers acknowledged an attempted fix and announced another change on November 6. The issue remains marked for verification. This directly matters to your laptop use. Issue #12
• November 4: a Codex app-server timeout report. Issue #8
• November 11: commit-message generation timing out on large diffs. Issue #32

These identify specific failures, not broader user sentiment. Independent long-term reviews are sparse. Older comparison pages are already stale: ohmyharness’s October 18 assessment lists only macOS and no public repository.

Fit and trial

For this workflow, choose it if a Mac remains your main supervision station and you want agents beside code on your own servers. Avoid relying on it as your sole solution if seamless laptop switching, Android access, or unattended server-side orchestration is essential.

Teams gain common conventions and provider flexibility. However, the enterprise page does not establish shared live workspaces, centralized permissions, audit logs, or fleet administration. “Agent team” primarily describes collaborating AI sessions.

The main concern is uncertain operational continuity around an experimental remote workflow. Trial it with one Ubuntu server and a disposable repository: run Claude implementation and Codex review, execute existing checks, disconnect Wi-Fi, close the laptop, restart the app, and attempt access from laptop two. Check process survival, transcript recovery, approvals, duplicated actions, and credential placement. Phone control and remote-workspace compatibility should also be tested together.

Comparison facts, reflecting the documentation above: