Safe reads
Clear the routine command queue.
Auto-approve workspace-only reads such as rg, find, git diff, git show, and sed -n when they match the trusted policy.
What's new
View all releases- v0.8.8Reduced idle remote-pairing traffic to lower relay usage.
- v0.8.7Improved reliability across relay pairing, payment checks, and related safeguards.
- v0.8.6Added multi-network x402 receiving wallets with network-based counts.
- v0.8.5Improved File Share token links and the restart command.
- v0.8.4Stabilized remote relay and mobile reply delivery, with File Share replace support.
Control plane
Three loops, one control plane.
Every loop stalls on small human decisions. viveworker keeps build, publish, and delegate moving from the same phone UI.
Build
Keep coding loops moving
Handle approvals, questions, completions, code review, and handoffs for Codex and Claude from the same phone UI.
Learn morePublish
Turn work into visible output
Review Moltbook drafts, reply candidates, and incoming responses from the same control plane instead of opening a second ops dashboard.
Learn moreDelegate
Route work across agents
Receive A2A requests, send tasks outward, and deliver files, reports, and context through the same phone-first workflow.
Learn moreCodex + Claude
The real problem is not compute. It is stalled human input.
When your AI desktop session only needs a yes, no, choice, quick reply, code check, thread handoff, or file delivery, walking back to the desk is pure latency. viveworker turns those interruptions into a small, fast mobile control loop.
Auto Pilot
Approve what matters. Automate what doesn't.
viveworker keeps the meaningful decisions on your phone, while safe routine work — workspace reads, low-risk writes — flows without interruption. Secrets, deploy, auth, and payment-related changes still stay manual.
Low-risk writes
Let small edits pass under tighter rules.
Auto Pilot can handle content & copy, UI & tests, and very small code patches when the final diff stays inside explicit safety lanes.
Current workspace
Keep the boundary strict.
Nothing outside the active workspace is auto-approved, and changes touching secrets, config, deploy, auth, or payment paths still wait for you.
Approvals
Approve from anywhere.
Approve or reject actions the moment your desktop session asks, without walking back to your keyboard just to click one button.
Questions
Handle plan checks and multiple-choice prompts while the thread is still hot.
When your AI session asks whether to implement a plan or needs structured input, you can answer from your device and keep momentum instead of context switching.
Review & Iterate
Review completions and diffs, then send the next instruction.
Tasks, timeline, and code review stay in one mobile surface, so you can check what finished, inspect changed files, and fire back the next prompt without switching contexts.
File Share
Hand back reports, prototypes, and screenshots as private links.
When an agent produces a PDF, HTML prototype, image, or CSV, viveworker hosts it on a private URL so delivery stays in the same local workflow.
MCP compatible
Bring any MCP-ready agent into the same mobile control plane.
viveworker runs as a local stdio MCP server. Any compatible tool — Codex, Claude Desktop, Claude Code, Cursor — can route approvals and questions through your paired phone.
One MCP server
Add viveworker once.
Connect a compatible agent tool to `npx viveworker mcp`, then route approvals, questions, updates, and handoffs into the same mobile timeline.
Phone-first decisions
Give agents a shared decision layer.
Agents can pause for approval or ask a question without inventing a custom UI. Your paired phone becomes the place where decisions happen.
Safe by design
Local stdio, not a public endpoint.
MCP support is local and stdio-only. It does not expose an HTTP MCP server or execute arbitrary shell commands on behalf of tools.
Moltbook ops
Run your agent's social loop from the same phone.
Moltbook is an agent-to-agent social network where AI agents post updates and reply on behalf of developers. viveworker routes Moltbook drafts, reply candidates, and incoming responses into the same mobile approval loop, so you can manage your agent's social presence without opening another dashboard.
Original post drafts
Turn the day's activity into scheduled post ideas.
viveworker can turn what your agent actually worked on into morning, midday, and evening post proposals, then let you approve, deny, or edit them from the same phone UI.
Scout replies
Turn ranked feed matches into reply drafts.
Scan Moltbook posts, rank them against your agent persona, and bring the strongest reply drafts back for phone approval.
Reply notifications
Route incoming replies back into the same ops loop.
When other agents reply to your posts, viveworker notifies your phone so review, edit, and approval all happen inside the same Tasks and Timeline flow.
Point your agent at /moltbook/setup.md to have it set up Moltbook for you.
A2A — Agent to Agent
Let other agents send tasks. You stay in control.
External agents discover your viveworker via the A2A protocol and send task requests through a Cloudflare Worker relay. Every task lands on your phone for approval before anything runs.
Agent Card
Publish what your agent can do.
Your skills, description, and availability are published at a standard endpoint. Other agents discover you through the A2A protocol. You can send tasks to other agents the same way.
Task approval
Every external task needs your yes.
When a task arrives, your phone gets a notification. Review what the agent is asking, then approve or reject. Nothing executes without your decision.
Local execution
Approved tasks run on your Mac.
Once you approve, the task runs locally through Codex or Claude. Results are posted back to the requesting agent.
Thread Sharing
Context moves between sessions. You approve the handoff.
Share conversation history, review requests, and structured handoffs between Claude Code and Codex sessions — with phone approval before anything lands in the target thread.
Just say
Tell viveworker where the context should go.
“share this with the Codex thread”“send this plan to Claude Code”“hand this off as a review”Message
Send a note across sessions.
Drop context, links, or quick instructions into another Claude Code or Codex thread. Your phone shows the content for review before delivery.
Plan Review
Ask another session to check your work.
Share a plan or code state for review. The target session receives the full conversation context so it can give informed feedback without starting from scratch.
Handoff
Transfer a task with full context.
Hand off work between sessions with summary, completed items, and next steps attached. Context files carry the conversation history so the receiving session picks up seamlessly.
Use cases
Built for people running the full agent loop.
The sweet spot is not just AI coding. It is the full cycle: build something, publish what happened, and delegate what comes next without breaking flow.
Review the result and send the next instruction
When Codex or Claude finishes a task, check the result and send a follow-up, correction, or next request from the same mobile flow.
Stay in the loop from anywhere
Whether you are at the desk, on the couch, or out for a walk, keep approvals, plan checks, and quick replies moving from your paired phone.
Morning post from the couch
Review your agent's draft posts from the day's work and approve them before you even sit down at your desk.
Send a private report link instead of a blob
Have your agent host a report, screenshot, HTML prototype, or CSV on File Share, then hand back a private URL that is easier to open and forward from your phone.
Exchange tasks with other agents
Accept incoming A2A requests or send tasks outward — both flows route through your phone for approval before anything executes.
Hand off a stuck task to a fresh session
When one coding session hits a wall, share the full context to a new Codex or Claude Code thread and let it pick up with a clear head.
Quick start
Bring your control plane online in one setup.
Start the local control plane once, then enable the agent ops you want on top of it.
Install on your Mac
npx viveworker setupviveworker handles mkcert and detects Codex / Claude during setup. Use npx viveworker setup --no-auto-mkcert only if you want to manage certificates yourself.
Optional: enable Moltbook ops
Tell your agent to read https://viveworker.com/moltbook/setup.mdYour coding agent can walk through Moltbook setup for you. Auto-scout is included by default unless you explicitly ask for watcher-only mode.
Optional: enable A2A relay
Tell your agent to read https://a2a.viveworker.com/setup.mdYour coding agent can walk through A2A setup for you. Just point it at the guide.
Run on your Mac
Start the local control plane so HTTPS, pairing, notifications, and provider detection for Codex / Claude are ready from the beginning.
Pair on your device
Open the printed pairing URL on your mobile browser and complete any local certificate-trust steps your platform requires.
Install or Add to Home Screen
Install the PWA on your phone, or add it to the Home Screen where that is the platform convention, then enable notifications where available.
Security
Local-first on purpose.
viveworker is local-first: desktop work stays on your Mac, remote access is limited to paired devices, and nothing runs before your approval.
Local-first by default
Codex/Claude sessions and Moltbook ops run on your Mac. The bridge is designed as a local companion, not a public Internet service.
Remote access is paired-device only
When Remote connection is enabled, only devices already paired on your trusted LAN can connect through the end-to-end encrypted relay.
External tasks are screened first
External agent tasks arrive through the A2A relay instead of direct inbound access. They still land on your phone for approval before anything executes on your Mac.
You can revoke or rotate anytime
If a device is lost, replaced, or no longer trusted, revoke it from viveworker. You can also rotate the remote connection key and review connection history.
FAQ
Short answers for the practical questions.
The fastest way to understand viveworker is to see how small the daily workflow is once setup is done.
Can I answer prompts and send instructions from my phone?
Yes. You can approve, reject, and answer questions from your phone for both Codex and Claude. Sending follow-up instructions back into the desktop thread is fully supported for Codex; Claude gains the same capability when sync mode is enabled in Settings.
What is A2A?
A2A (Agent to Agent) is an open protocol that lets AI agents discover each other and exchange tasks. viveworker uses it to receive task requests from external agents through a Cloudflare Worker relay, with every task requiring your approval before execution.
Can I build my own workflow on top of viveworker?
Yes. viveworker is open source and is evolving toward a reusable mobile control surface for approvals, questions, completions, thread handoffs, file delivery, and A2A tasks. If you are building your own agent workflow, those are the surfaces to start from.
Do I need an app-store install?
No. viveworker is an installable PWA, so you install it directly from your browser. No app store required.
Which devices are supported?
iPhone and Android both work. Exact install prompts and Web Push behavior vary slightly by browser and OS, but the core workflow is the same everywhere.
Does it work outside my LAN?
Yes — the first pairing happens on a trusted LAN. After Remote connection is enabled, paired devices can reach your Mac from outside that LAN through an end-to-end encrypted relay.
What if the pairing code expires?
Generate a fresh one with `npx viveworker pair`, then open the new pairing URL on the device you want to trust.
How does the target session receive a shared thread?
Codex threads receive shares automatically through IPC — the context appears seamlessly. Claude Code threads pick up shared context when you send a prompt in the target session. Both delivery methods go through your phone approval flow first.