Ask what is wrong. RemoteX checks your servers, clusters, databases, logs, files, and Docker apps, shows the evidence, proposes a fix, and runs only what you approve.
Use it from your phone when production breaks away from your desk, or on desktop when you want room. Direct connections. No agent to install. No RemoteX proxy.
Mac, Linux, Windows, iPhone, and Android. Desktop is live
Bring your own AI. Works with
14 providers. One key, one sign-in, or your local model. Your device talks to the provider you chose. RemoteX never proxies and never sees your prompts.
Now on Mac, Linux & Windows
RemoteX pulls the investigation into one place: the question you asked, the servers involved, the command output, the files, the database context, and the next approved fix.
Checking uptime, disk, memory pressure, listening ports, Docker containers, and database processes across the attached servers.
sh -lc 'uptime; echo ---; df -h /; echo ---; docker ps | head' load average: 0.06, 0.04, 0.03 /dev/nvme0n1p2 292G 105G 177G 39% / postgres mariadb mysqld redis-server api Up 3 days · worker Up 3 days …
The server is healthy. No immediate resource issue is evident; containers are running and the database services are not consuming excessive CPU or memory.
Same agent loop and safety classifier across desktop and mobile, sharing a single core.
One workspace
When production is noisy, the hard part is not opening another terminal. It is keeping the server, cluster, database, logs, files, and monitoring story in your head. RemoteX keeps them attached to one thread, with one approval model.
Projects group the infrastructure that belongs together. Add SSH servers, Docker hosts, Kubernetes clusters, Postgres/MySQL/SQLite/Redis connections, and approved local folders to the same thread.
The agent can reason across attached context while every command, kubectl action, SQL mutation, or file write still names its target and waits for approval.
The workflows below mirror the real app: let the agent use its own terminal, save repeat investigations, turn plain-language alerts into rules, and keep the answers you will need again.
Every command the agent runs lands as $ command followed by the actual output beneath it, in pure monospace flow — no chat noise. Type your own shell commands or SQL queries into the green ❯ prompt at the bottom: they execute in ~100ms with no LLM round-trip, and the agent reads them as context on your next turn. So if you debug something by hand, then ask the agent a follow-up, it already knows what you saw.
Add a database the same way you add an SSH server. The right pane gains a Results tab next to Activity: schemas + tables tree on the left, one click loads a sample of any table as a sortable, searchable view — columns, types, NULL handling, JSON cells, the works. Long string cells expand to a modal. Hit the maximize button and the whole thing pops to a TablePlus-style full-window explorer.
The agent reads the same connection. Ask "what changed in users this week?" and it queries with row caps + read-only guards, results stream back into the table.
$-mention from chat.Save the things you do often. Skills are reusable instructions and intents, including file-backed SKILL.md from your project or personal agent setup. Workflows are multi-step routines with approval gates. Type $ in chat to splice either into your message.
Tell your AI provider what you want — "warn me if disk fills above 85% for ten minutes" — and it compiles a deterministic rule the app evaluates locally, forever. One LLM call at authoring time, then zero ongoing cost. Same idea for workflows: describe the steps, the agent writes the runbook.
Turn on Plan mode when you want the agent to propose the investigation first and wait for approval. Once approved, the same checklist stays on screen and ticks forward as each step starts and finishes.
/summarize gives you a quick recap when you just need to hand off or close the loop. /memorize stores the useful conversation verbatim, on-device and per server, so a later chat can recall the exact context instead of guessing.
Memory is local-only and included: RemoteX recalls what matches the current question, then still verifies anything time-sensitive before acting.
Drop a server file or folder from the desktop right rail into the composer. RemoteX carries the server id, file kind, and full remote path into the next turn, so "what is this file?" means the thing you dragged, not a guessed filename from the conversation.
Attach a skill or workflow to any rule. When the rule fires, the local notification body suggests the linked action — one tap takes you from the alarm into a pre-seeded investigation chat on the right server, with the agent already mid-diagnosis.
CPU, memory, disk, load, and failed services per host, each with a sparkline you can scrub from 1 hour to 30 days. Alert rules live right under each card, so you see the metric, its trend, and the thresholds protecting it in a single glance.
Star any message — the agent's synthesis, the exact command that fixed it, your own clever follow-up — and find it in a Pinned view spanning every server. The investigation you ran on Tuesday is one tap away the next time the same alarm fires.
Sonnet for everyday server work, GPT-5 for a second opinion, Gemini for long context — switch the whole picker mid-thread, no context lost.
The composer also keeps the safety mode and active server in reach, so the setup for a new thread stays in one quiet bar.
A tour
A disk fills up, a pod crash-loops, a query slows down, an alert fires while you are away from the desk. These previews show the workflows, not just the feature names.
Every state-changing command surfaces as a card with the exact bytes about to run, a short explanation, and an explicit risk label. Face ID on iPhone, Touch ID on Mac, every time — even if you typed it yourself.
The agent runs read-only diagnostics autonomously and streams every line as it arrives. You see the exact same evidence the AI is using, no black box, no "trust me" summaries.
One sidebar holds your full fleet. Each server expands to show its recent investigations as threaded chats. ⌘K on desktop jumps to any thread, skill, or workflow across servers.
RemoteX samples each server directly over SSH from your device. Load, CPU, memory, disk, failed services. No agents to install, no extra dashboards to babysit, no proxy infrastructure required.
On Mac, RemoteX detects ssh-agent and walks ~/.ssh/id_ed25519, id_ecdsa, id_rsa on the first add screen. Pick one with a click instead of pasting a private key blob.
Use your ChatGPT subscription or GitHub Copilot seat instead of generating an API key. RemoteX never proxies the call; tokens live in your Keychain and traffic goes straight to OpenAI or GitHub.
Walk a server's filesystem like an IDE — collapsible tree, syntax-highlighted preview, read-only by default. And point RemoteX at a local folder you approve: the agent reads your own code project too, so it diagnoses the server with the full picture, not half of it. Open nginx.conf on the box and the matching config in your repo, without leaving the chat.
Servers, Kubernetes clusters, database connections, chat threads, messages, and rendered tool outputs sync between Mac, iPhone, and iPad through your own iCloud Drive container. A passphrase only you know is the key — Apple sees a sealed blob, RemoteX has no key, no recovery, no copy. No central server, no account. Add a server on your Mac and see it on your phone within a beat of focusing the screen.
Export every server, AI key, skill, workflow, and thread into a single .rxbackup file. Encrypted with scrypt + XChaCha20-Poly1305 and your passphrase. Open it on the other device — phone or Mac, and everything restores in seconds. Complements iCloud sync for offline archival and bringing a brand-new device online without an existing iCloud workspace.
Linux hosts, Kubernetes clusters, and Postgres / MySQL / Redis / SQLite live in the same Connections list. The agent picks the right tools for the kind — shell on SSH, kubectl on K8s, read-only SQL on databases — and a single chat can talk to several at once. Databases reached through an SSH bastion just work.
Attach a file from your phone or Mac and RemoteX shows it as a labelled chip in the composer. PDFs read natively on Claude and Gemini; spreadsheets, DOCX, PPTX, CSV, logs, and code are extracted on-device. The model sees the contents alongside your question, no copy-paste required.
One workspace
How it works
Your device connects directly to your infrastructure and to the AI provider you choose. RemoteX gathers the evidence, proposes the next step, waits for approval when state changes, and helps verify the result.
Add SSH servers, Docker hosts, Kubernetes clusters (paste a kubeconfig), and Postgres / MySQL / SQLite / Redis. Bastion-tunneled targets work too. Credentials stay in the OS secure store.
Plain language — site down, disk full, slow query, deploy failed — or attach a screenshot, log file, or kubeconfig for context.
RemoteX runs read-only checks: reads logs and files, samples rows or query plans where allowed, lists pods and describes resources, and builds a diagnosis while you watch.
Any state-changing action is shown first, in plain bytes, and requires biometric or local approval. Every time.
After the change, RemoteX re-checks services, logs, ports, query results, and responses so you can confirm it worked.
RemoteX does not proxy infrastructure traffic. SSH, Kubernetes, and database connections go straight from your device to your target. Prompts go straight to the AI provider you chose. Only limited billing/license metadata ever touches a RemoteX service.
Security
AI infrastructure tooling only works if you can trust the path. RemoteX keeps credentials on-device, connects directly to your targets, sends prompts only to the provider you chose, and asks before anything changes.
SSH keys, kubeconfigs, database passwords, and AI provider keys live in iOS Keychain, Android Keystore, macOS Keychain, libsecret on Linux, or Windows Credential Vault. They are never transmitted to a RemoteX server.
SSH, Kubernetes, and database connections run direct from your device to your infrastructure. AI prompts go directly to the provider you chose. No analytics SDK. No RemoteX-operated cloud holds your servers, commands, queries, or conversations — optional cross-device sync runs through your own iCloud Drive, end-to-end encrypted with a passphrase only you know.
Read-only checks run autonomously while you watch. Anything that mutates state — rm, systemctl restart, kubectl apply, UPDATE / DELETE SQL, package installs, file writes — is shown first and requires biometric (Face ID, Touch ID) or local-passphrase confirmation.
Desktop can re-lock after inactivity behind a local PIN or password gate. Credentials stay in the OS secure store; an idle workspace doesn't sit open on your desk.
14 providers including run-it-yourself Ollama and LM Studio, plus a generic OpenAI-compatible slot. Bring your own key or sign in with ChatGPT or GitHub Copilot. Usage is billed by the account that owns the key.
FAQ
A local-first AI workspace for remote infrastructure operations. It connects to your SSH servers, Kubernetes clusters, and databases directly, helps you inspect and diagnose problems, proposes fixes, and only changes state when you approve. Desktop and mobile, sharing one core.
On your device, in the OS secure store: iOS Keychain, Android Keystore, macOS Keychain, libsecret on Linux, or Windows Credential Vault. RemoteX never sees them.
No. SSH, Kubernetes API, and database connections go directly from your device to your infrastructure. AI prompts go directly from your device to the provider you chose. RemoteX operates billing/license services for Pro purchases, trusted-device limits, and pairing — and never receives your server list, credentials, commands, output, queries, or conversations.
14 of them. Anthropic, OpenAI, Google Gemini, OpenRouter, xAI, DeepSeek, Mistral, Vercel AI Gateway, Vercel v0, Amazon Bedrock, GitHub Copilot, plus run-it-yourself Ollama and LM Studio, plus a generic OpenAI-compatible slot for anything else with an OpenAI-shaped API. Bring your own API key, or sign in with your ChatGPT or GitHub Copilot subscription where the provider supports OAuth. Usage is billed by the provider account that owns the key or seat.
Only read-only commands can run autonomously. Anything that could change state requires Face ID, Touch ID, or your local passphrase (Linux / Windows) before it runs, every time, no exceptions, even if you initiated the request yourself. After an approved fix, RemoteX runs verification checks so you know whether it actually worked.
Per-server rules sample CPU, memory, disk, load, and failed services directly over SSH — no agent installed on the server. A local notification fires when a rule breaches, and you can attach a skill or workflow so one tap drops you into a pre-seeded investigation chat. Desktop runs continuous checks while open; mobile is best-effort within iOS / Android background limits.
Yes — three. The desktop app is an Electron build with projects, multi-server chat context, Kubernetes support, database connections, ⌘K command palette, biometric or local approval, SFTP file browser with syntax highlighting, terminal access, local app lock, and silent auto-update. It runs the same agent loop as mobile through a shared core package. The Mac DMG is Apple Silicon and notarized; Linux ships as an AppImage (every distro) and a .deb (Debian, Ubuntu, Mint), both x86_64; Windows ships as an x64 NSIS installer plus a portable .exe (unsigned preview — SmartScreen may warn on first run).
Yes, two ways. iCloud sync handles ongoing cross-device use between Mac, iPhone, and iPad: servers, chats, messages, and rendered tool outputs sync end-to-end encrypted through your own iCloud Drive container, with a passphrase only you know. Apple sees a sealed blob, RemoteX has no key, no recovery, no copy. Sync uses per-record CRDT merging so two devices can edit different things at once and both edits survive. Encrypted .rxbackup files (Pro) cover one-shot migration and offline archival: a single file containing servers, AI keys, skills, workflows, monitoring rules, and threads, encrypted with scrypt + XChaCha20-Poly1305 and your passphrase. Restore on any platform.
Yes, on iOS, Android, Mac, Linux, and Windows. Long-press the composer mic and dictate the question; the agent picks it up like any other prompt. iOS uses on-device recognition where available, so the audio doesn't leave the phone. The desktop apps use the system Web Speech API.
Free covers the full workspace on one connection, including skills, workflows, monitoring rules, memory, and AI chat with any supported bring-your-own API key. Pro unlocks unlimited connections plus encrypted backups with credentials. One subscription works on iOS, Android, Mac, Linux, and Windows.
iOS is live on the App Store. Android is live on Google Play. Mac, Linux, and Windows builds live on the desktop page (or jump straight to /linux or /windows).
Download
iPhone · Android · macOS · Linux · Windows preview. One subscription covers all of them.
Free forever for the full AI agent on one server. See Free vs Pro →
