OpenTandem
Run and steer the 18 AI coding agents on your own computer from your phone, with one shared workspace.
- Source repo
- opentandem/opentandem
- Stars
- ★ 14
- Last updated
- 16d ago
- License
- AGPL-3.0
- Primary language
- JavaScript
- FA score
- 44/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificCodex · Claude CodeChatGPT (Partial support)
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer running Claude Code or Codex on a home laptop who wants to keep the same task moving from a phone while away.
- Not a fit if
- Teams that need fully offline operation and cannot send task data to a hosted workspace
- Solo users who want a single local model and no existing agent CLI installed
- Teams unwilling to use a hosted collaboration workspace and needing an all-self-hosted stack
- Website
- opentandem.org ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-05
- Source review
- 44/100 · Major gaps
What does this agent do, and when should you use it?
OpenTandem is a workspace for individuals and teams who run several AI coding agents on their own machines and want to drive them from a phone. Two pieces are open source under AGPL-3.0: the opentandem CLI, which pairs a computer to the workspace, keeps a background service alive, starts it at login and self-updates, and the Runner, which launches the agent CLI you pick inside your project folder and streams progress, answers and artifacts back. The workspace itself — tasks, conversations, machines, teams and mobile access — is the hosted service at opentandem.org. Setup is one command on the machine that already has the agents, a QR scan with your phone, and an Allow tap; after that you choose a machine, a project folder and an agent from the app. All 18 listed agents are supported, including Claude Code, Codex, OpenCode and Gemini CLI, and you can switch agents and models mid-task while keeping the conversation and outputs.
The CLI registers the host as a Runner via npx opentandem@latest remote (or the install.sh / install.ps1 scripts), prints a QR code and a pairing link valid for 10 minutes, and waits for the user to sign in on a phone and tap Allow. It then runs as a background service that starts with the computer; opentandem list shows each local Runner id and online state, start/stop/restart take an id or all, logs [-f] follows output, update refreshes CLI and Runner, and uninstall removes the service while keeping local conversation history. When a task arrives, the Runner starts the chosen agent CLI — already installed and signed in with the user's own account — inside the selected project folder, streaming execution progress, answers and generated images, documents and HTML pages back to the phone or web app. The workspace organises tasks by project, saves, archives and pins conversations, and surfaces live Runner availability and per-agent quota beside the composer.
- A developer running Claude Code or Codex on a home laptop who wants to keep the same task moving from a phone while away.
- Someone with several machines (a Mac mini, an SSH-reachable Linux server) who needs to see which one is ready before assigning work.
- A team comparing agents or models inside one conversation and switching them without rebuilding context or losing outputs.
- Teammates sharing a project task so a colleague can continue the same execution instead of starting a fresh thread.
- Long-running agent work on a fixed workstation, reviewed by opening artifacts and HTML previews from a phone.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx opentandem@latest remoteirm https://opentandem.org/install.ps1 | iex; opentandem remote
- Install notes cover
- Windows · macOS · Linux
How do you install or deploy this agent?
On the computer where your coding agents already live (requires npx):
npx opentandem@latest remoteWithout Node.js, use the installer on macOS and Linux:
curl -fsSL https://opentandem.org/install.sh -o install.sh
sh install.sh remoteOn Windows PowerShell:
irm https://opentandem.org/install.ps1 | iex; opentandem remotePrerequisite: at least one supported agent CLI (Claude Code, Codex, OpenCode, and so on) is installed and signed in on that machine.
How do you use this agent?
- The command prints a QR code and a pairing link (e.g. https://opentandem.org/agent-pair/...) valid for 10 minutes.
- Scan it with your phone, sign in with your email and tap Allow to connect the computer; the CLI confirms the connection and runs in the background.
opentandem remote
- In the OpenTandem app or web workspace, pick this computer, a project folder and an agent, then send the task; the agent runs locally and progress and outputs are viewable on the phone.
- Day-to-day commands:
opentandem list
opentandem stop <id>
opentandem start all
opentandem logs -f
opentandem update
opentandem uninstall
An AI agent may run npx opentandem@latest remote for you, but only you can scan, sign in and allow the computer.
What are this agent's strengths and limitations?
- Onboarding is one npx command plus a QR scan, with no server-side deployment work.
- Code stays in local project folders and runs through agent CLIs you already signed in with, so repos and credentials are not migrated to a cloud IDE.
- You can switch among all 18 supported agents and change models within one task while keeping the conversation and prior outputs.
- Runner-level live availability, log following, start-at-login and self-update make it practical for always-on machines.
- The CLI and Runner are AGPL-3.0 open source, and the README draws a clear line between those parts and the hosted service.
- Tasks, conversations and team features depend on the hosted opentandem.org service; the repository does not provide a self-hostable backend.
- You must already have and be signed in to a third-party agent CLI, adding another vendor and account dependency.
- First connection requires a phone scan and sign-in, so it cannot be fully scripted or run unattended.
- Collaboration and multi-machine orchestration are built around the vendor cloud, which blocks teams whose task data must stay internal.
- Pull requests are not accepted yet, so deep customisation means maintaining your own fork.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OpenTandem This agent | 44 · Major gaps | CLIFreemium | ★ 14 | 16d ago | JavaScript | Codex · Claude Code |
| Ordewell | 87 · Good | CLIFree + model costs | ★ 211 | 1d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| Postbag | 78 · Good | — | ★ 10 | 1mo ago | Python | Codex · Claude Code |
| Agetor | 76 · Good | — | ★ 137 | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The plugin capability layer treats each capability name as a permission and marks writes as needing member confirmation (capabilities/index.js), and the server client keeps the secret server-side — positive least-privilege and confirmation evidence. However the Runner launches 18 agent CLIs on the user's machine and executes arbitrary tasks; the README does not state those agents' permission scope, file-access boundaries, or network egress, and data flow is only a high-level mermaid diagram without saying which conversations/files reach the hosted server. The update path has an allow-list and version-string validation (agent-update-runner.test.ts), but the README states every run auto-updates to latest first — an unconfirmed automatic external change — and rollback is limited to uninstall preserving local history, with no version-pinning or revert mechanism. No lockfile or dependency-audit evidence; source attribution rests on README and AGPL text only, and the publisher is unverified.
Tests show fairly detailed unit coverage of install detection, update planning, concurrent slots and a memory gate, and internal naming stays consistent (CAPABILITY_NAMES kept in step with the server per comments), so self-consistency is decent. Dependency availability (npx fetch, curl|sh installer, hosted server) has no version pinning or offline story, and failure messaging is only structured in the server client (OpenTandemError); Runner/CLI failure text is not shown in the sample.
The README clearly targets individuals and teams, multi-machine and mobile scenarios, and covers macOS/Linux/Windows install paths, so environment fit is well described. Capability boundaries (what agents may and may not do) are undefined, trigger precision depends on the user manually picking an agent and project with no routing rules documented, and the sample lacks SKILL/plugin manifests to bound triggers.
The README is well structured with a command table, architecture diagram, parts table, contributing and security-reporting channels, and complete install notes (npx, install.sh, PowerShell). But there is no CHANGELOG file (only an external site link), no versioning policy, no known-limitations section, naming stability cannot be judged from the sample, maintenance responsibility points only to GitHub issues with PRs explicitly not accepted yet, and publisher identity is unverified.
The value proposition is clear (phone remote control of local agents, multi-machine orchestration, team collaboration) and output usability is supported by GIFs and a command table. Marginal value depends on the hosted service, and the cost side (auto-update, background daemon, data through the server) versus benefit is not quantified; the sample cannot verify actual results.
Some README claims are partly corroborated by tests (install detection, update allow-list), but most product-level claims (18 agents supported, team collaboration, quota tracking) come only from the README and media with no matching source or tests; fact and inference are not clearly separated in the docs, and external links point to the hosted site rather than checkable evidence.
- The README states every run auto-updates to latest first — an external change without user confirmation — with no documented rollback or version pinning.
- The Runner launches 18 agent CLIs and executes tasks on the user's machine, but file-access scope, network egress and permission boundaries of those agents are not described.
- Data flow is only a high-level diagram; it does not say which conversations, files or credentials are sent to the hosted opentandem.org server.
- Installation relies on curl|sh and npx pulling latest, with no lockfile or integrity-check evidence, leaving supply-chain risk unmitigated.
- Publisher identity is unverified; maintenance responsibility and update path point only to GitHub issues, and PRs are not accepted yet.
- The sample lacks a CHANGELOG, known-limitations section and versioning policy, so naming and interface stability cannot be judged.
FAQ
Does it upload my source code?
Do I need Node.js?
npx opentandem@latest remote needs Node.js. Without it, use install.sh on macOS/Linux or install.ps1 in Windows PowerShell.Can I self-host the whole thing?
Which agents are supported?
What happens if I do not complete pairing?
opentandem remote to get a fresh one. Signing in and allowing the computer can only be done by the user on the phone.Is OpenTandem free?
Does OpenTandem work on Windows?
irm https://opentandem.org/install.ps1 | iex; opentandem remote.Is OpenTandem open source?
Is OpenTandem safe to use?
Is OpenTandem still maintained?
Who makes OpenTandem?
What is OpenTandem's official website?
Thanks — we use this to improve the page.