Harness Remote
Control and continue native coding-agent sessions across your machines from desktop, web, or Android.
- Source repo
- giuliastro/harness-remote
- Stars
- ★ 423
- Last updated
- today
- License
- Apache-2.0
- Primary language
- JavaScript
- FA score
- 69/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer starts a Codex CLI session on a home workstation and later checks its progress or continues from an Android device.
- Not a fit if
- Teams that need to expose the gateway directly to the public internet
- Teams that need the tool itself to sandbox coding-agent permissions
- Source review
- 69/100 · Some gaps
What does this agent do, and when should you use it?
Harness Remote is a local-first control plane around the coding agents you already run; it does not replace them or invent a universal conversation model. A Machine gateway discovers and presents your machines, Projects, and native Sessions for Codex CLI, Claude Code, GitHub Copilot CLI, OpenCode, MiMo Code, Oh My Pi, and PI. From desktop, web, or Android, you can follow activity, send prompts, stop turns, and handle questions or permissions supported by the harness. Continuation creates a real native Session on the target harness and records its relationship to the source; work can also continue on another configured machine. Code, credentials, subscriptions, and native Session persistence stay on your machines, and remote gateways are intended for a trusted LAN or VPN.
After at least one supported coding-agent CLI is installed and authenticated on the machine to control, run npx harness-remote to start one Machine gateway; it detects installed agents, chooses available ports, and generates credentials. Harness Remote presents Machines → Projects → native Sessions, where clients can observe activity, send prompts, stop turns, and surface questions or permissions the agent supports. Codex takes a prompt received mid-turn into the running turn; the other documented harnesses queue prompts for later handling or withdrawal. Cross-agent or cross-machine continuation creates a native Session on the target harness and records the relationship; bounded context can include the objective, decisions, unresolved work, and checks already run. The coding agent remains responsible for conversation semantics, tool execution, model behavior, and resume semantics.
- A developer starts a Codex CLI session on a home workstation and later checks its progress or continues from an Android device.
- A teammate needs to observe native coding-agent sessions on a development machine over a trusted LAN or VPN.
- An OpenCode user wants to hand a task to Codex or Claude Code while keeping a recorded link to the source Session.
- A developer starts work in a normal CLI and later wants to find that same native Session in the desktop or web client.
- A user working across configured machines wants to continue a project while preserving Project identity and Session lineage.
How do you install or deploy this agent?
For desktop, download the latest GitHub release and open Harness Remote; on Windows and macOS, the app automatically starts and supervises its local Machine runtime. To add another machine, install Node.js 20+ and ensure at least one supported coding-agent CLI is installed and authenticated, then run:
npx harness-remoteA direct-from-GitHub option is:
npx --yes github:giuliastro/harness-remoteTo connect the hosted web client, allow its origin in the gateway's CORS configuration:
npx harness-remote --cors https://giuliastro.github.ioAndroid pairs by scanning a short-lived QR code generated by the gateway or app. The README does not state the software or coding-agent CLI pricing.
How do you use this agent?
Keep the gateway terminal running on the machine you want to control. Optional settings can restrict Project selection, choose a port, or allow a web origin:
npx harness-remote --root ~/dev
npx harness-remote --port 4900
npx harness-remote --cors https://giuliastro.github.ioIn a client, open Machines, choose a machine and Project, then open a native Session. On Android, select Scan machine QR code and scan the gateway's QR; from desktop or web, use Pair a phone on the Machines page to generate one. Use remote gateways only over a trusted LAN or VPN; do not expose them directly to the public internet. --root limits Project selection and is not an operating-system sandbox.
What are this agent's strengths and limitations?
- Keeps each supported tool's native Session, context, permissions, and resume semantics authoritative.
- Continues work across agents and configured machines while recording Session relationships.
- Offers desktop, web, and Android access; the Windows and macOS desktop app manages its local runtime automatically.
- Discovers models and capabilities from the running harness instead of hardcoding controls for every agent.
- Requires Node.js 20+ and at least one supported coding-agent CLI installed and authenticated.
- The gateway must keep running; stopping or restarting it interrupts any turn still in progress.
- The documented deployment is for a trusted LAN or VPN; the gateway should not be exposed directly to the public internet.
--rootis not a security sandbox, and coding agents retain the permissions of the account that launched them.- Transcript, tool execution, model behavior, and resume semantics remain dependent on each harness.
How does this agent compare with similar options?
Compared with using a coding-agent CLI alone, Harness Remote adds machine, project, and session discovery, remote control, and cross-agent or cross-machine continuation around it. The selected agent still owns execution, context, and its native Session.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Harness Remote This agent | 69 · Some gaps | Desktop app | ★ 423 | today | JavaScript | Codex · Claude Code |
| Portable Handoff | 84 · Good | CLIFree | ★ 22 | 1mo ago | Python | Codex · Claude Code |
| Happier | 82 · Good | Desktop appFree + model costs | ★ 2k | today | TypeScript | Codex · Claude Code |
| Caprock | 62 · Some gaps | Desktop appFreemium | ★ 25 | today | Go | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README says code, credentials, and native Sessions remain on the user's machines, recommends trusted LAN/VPN use, and states that --root is not an OS sandbox. ACP tests show permissions are denied by default, can be granted once, and desktop phone pairing requires confirmation. Deductions: agents still run with the launching account's permissions, the gateway enables remote prompts and control, dependency-security evidence is limited, and no clear rollback mechanism is shown; publisher identity is unknown and the files do not identify a maintainer.
ACP client tests cover initialization, concurrent start, pagination, timeouts, process exits, EPIPE, and actionable error details; the README says a gateway restart interrupts an active turn. Deductions: test files do not establish executed results, and the package lists only a small set of direct dependencies with limited static evidence about availability or recovery.
The README describes desktop, Web/PWA, Android, multi-machine use, and integrations with several native agents; it points to a capability matrix and explains native Session ownership and environment prerequisites. Deductions: trigger conditions are spread across documents, and the supplied files do not fully establish environment requirements for every agent.
The README organizes quick start, features, security, development, and reference material; it gives Node.js 20+ instructions and command examples. Version 3.5.0 is declared and the Apache-2.0 license text is present. Deductions: the supplied material contains no changelog, FAQ coverage is limited, maintenance ownership is unclear, and installation detail varies by platform and agent.
The product focuses on remote observation, native Session recovery, and continuing work across agents or machines. It provides inspectable handoff context, Session status, and diagnostics while preserving native agent semantics. Deductions: users must deploy and maintain a local gateway, and multi-machine and phone-pairing features add setup cost; the files do not quantify savings or reliability gains.
Core product boundaries and installation claims can be compared across the README, package metadata, license, workflows, and ACP tests; tests describe specific client behavior and error handling. Deductions: many feature claims remain README assertions, the implementation source and full referenced documentation are not supplied for item-by-item review, and static evidence cannot verify runtime behavior, so confidence remains low.
- Do not expose the gateway directly to the public internet; the README recommends using a trusted LAN or VPN.
- --root only limits project selection; it is not a sandbox, and agents inherit the launching account's permissions.
FAQ
Does Harness Remote cost money?
Does it send my code or credentials to a Harness Remote cloud?
What happens if the gateway stops or disconnects?
Can it restrict what files a coding agent can access?
--root limits which directories Harness Remote offers for Project selection, but it is not an operating-system sandbox. The coding agent still has the permissions of the account that launched it.