amux for Freelancers

Run an AI engineering team from one dashboard. Bill hours while agents do the coding, manage every client relationship in one place, and deliver completed features overnight. Updated July 2026.

amux is an open-source control plane that gives freelance developers a parallel AI engineering team — without hiring. It launches, monitors, and self-heals simultaneous Claude Code, Codex, or Gemini CLI sessions across your client projects, with a built-in kanban board, notes, CRM, Gmail API integration, browser automation, and a scheduler. One Python file, zero external dependencies, runs on your MacBook, and is controllable from your iPhone at a client meeting.

How freelancers use amux to deliver more while working less

The ceiling on freelance income is usually hours. There are only so many billable hours in a week, and adding more clients means working longer — or turning work down. Most freelancers plateau at 2-3 clients because context-switching alone eats 20% of their available time.

AI coding agents change the arithmetic. With amux, you can run one dedicated agent per client project simultaneously. While you're on a call with Client A, agents are writing code for Clients B, C, and D. While you're sleeping, agents are completing features queued that afternoon. When you wake up, you review PRs, do a quick QA pass, and deliver. The bottleneck shifts from "how many hours can I personally code" to "how fast can I review and communicate."

The practical result: freelancers using amux report taking on 3-5x their previous client load without increasing working hours — and often reducing them. The agents do the execution; you do the strategy, communication, and quality control.

Run multiple client projects in parallel

The core amux pattern for freelancers is one session per client project. Each session is isolated: its own working directory, its own git worktree, its own CLAUDE.md with that client's coding conventions, linting rules, and project context. Agents never touch files outside their registered directory.

# Register each client as a separate session
amux register client-acme --dir ~/Dev/acme-app --yolo
amux register client-beta --dir ~/Dev/beta-platform --yolo
amux register client-gamma --dir ~/Dev/gamma-api --yolo

# Start all three in parallel — agents work independently
amux start client-acme
amux start client-beta
amux start client-gamma

# Queue tasks per client on the board
amux board add "Implement CSV export for reports page" --session client-acme
amux board add "Fix checkout flow redirect bug" --session client-beta
amux board add "Add rate limiting to /api/search endpoint" --session client-gamma

The amux dashboard at https://localhost:8822 shows all sessions in a single view — live output, board status, agent health — so you can see what every client's agent is doing at a glance. The mobile PWA gives you the same view from your phone.

Agents follow the CLAUDE.md style guide in each project root, so code for Client A follows Client A's conventions and code for Client B follows theirs. Quality is consistent because the agent is always working from the same written spec you gave it — not guessing from context.

The CRM is built in — manage clients where you manage agents

Most freelancers keep client contacts scattered across their phone, email, Notion, and memory. amux includes a lightweight CRM at /api/crm/contacts — contacts with company, email, role, notes, interaction history, and follow-up tracking. It lives in the same tool that runs your agents, so you never context-switch to a separate app to log a call or find a client's email.

# Add a client contact
amux crm add "Sarah Chen" company="Acme Corp" email="[email protected]" role="CTO"

# Log an interaction after a call
amux crm log PPL-1 "Reviewed Q3 roadmap. She wants the CSV export by Friday."

# Check pending follow-ups across all clients
amux crm fu

The Gmail API integration lets you send client updates directly from the amux CLI or dashboard — with your real Gmail signature attached, proper threading, and CC support. You can send a "here's what the agent completed this week" summary without opening a browser.

# Send a client update from the CLI
curl -sk -X POST -H 'Content-Type: application/json' \
  -d '{"to":"[email protected]","subject":"Week 28 delivery","body":"Hi Sarah,\n\nAgents completed the CSV export and fixed the mobile layout bug. PRs are open for your review:\n- github.com/acme/app/pull/142\n- github.com/acme/app/pull/143\n\nReady to merge when approved.\n\nEthan","from":"[email protected]"}' \
  $AMUX_URL/api/email/send

The scheduler can automate these updates: send a weekly delivery summary every Friday at 5pm, log a follow-up reminder if a client hasn't responded to a PR in 48 hours, or ping yourself when an overnight run completes. For details, see the getting started guide.

Overnight delivery: queue work before bed, review in the morning

The highest-leverage freelancer pattern in amux is the overnight run. Here's how it works in practice:

1. End-of-day: write atomic task cards. Before you close the laptop, turn the day's client conversations into specific task cards on the board. "Add CSV export to reports — output format matches the attached mockup, columns: date, user, event, value" is a good task. "Improve the analytics page" is not. Agents need specificity to succeed unattended.
2. Enable YOLO mode. With --yolo, agents run without stopping for permission prompts. Essential for overnight unattended work. See the AI coding while you sleep guide for safe YOLO configuration.
3. Set a CLAUDE.md per project. Each client project should have a CLAUDE.md in the root with coding standards, project context, gotchas, and style preferences. Agents read this before starting any task — it eliminates the most common failure mode (agent uses wrong patterns).
4. Agents open PRs, not direct commits. Configure each session to commit to a branch and open a PR rather than pushing to main. You review in the morning with a clear diff, approve or leave comments, and the agent can address feedback in round 2.
5. Morning review from your phone. The amux iOS app shows which board tasks moved to "done" overnight, with links to the PRs. You can review a diff, leave a comment, and approve a merge from the phone before you get out of bed.

A realistic outcome for a well-prepared overnight 3-agent run across 3 client projects: 2-4 tasks completed per project, 6-12 PRs total ready for review by 7am. Compare that to the 1-2 tasks you might complete in a full day of focused solo coding.

Check from anywhere — mobile-first while at client meetings

The amux dashboard is a mobile PWA designed to be useful from a phone, not just a laptop. When you're at a client meeting or a coffee shop, you can check agent status, read board updates, send a quick correction message to an agent that went sideways, or trigger a new task — all without opening a computer.

Common in-the-field actions via the iOS app:

• See which client agents are running vs. idle
• Peek at the last 50 lines of output from any session
• Move a task from "todo" to "doing" or "done" with a tap
• Send a clarifying message to a session ("use the existing formatCurrency() helper, don't write a new one")
• Trigger a new task for an idle agent from a requirement that just came up in the meeting

The session-peek feature is particularly useful during client calls — you can genuinely say "let me check where the agent is on that" and pull up live output right there. It sets a strong impression that work is actively in progress (because it is).

How much can a freelancer realistically scale?

The question every freelancer asks before committing to this setup. The honest answer depends on task quality and review discipline, but here are realistic numbers:

The limiting factor at scale is not agent capacity — it is your review throughput. A freelancer who can review 8-10 PRs per day clearly, provide precise feedback, and re-queue follow-up tasks well can sustain 4-6 concurrent client relationships. Beyond that, the review queue becomes the bottleneck. The practical ceiling for a solo freelancer running amux is typically 5-6 active clients at premium billing rates — roughly 3-4x the solo maximum.

For agent cost estimates, see the AI coding agent cost guide. Short version: Claude Max at $200/month covers unlimited overnight runs; API pricing runs $10-30/night for a 5-agent run on Claude Sonnet.

Freelancer workflow: with amux vs. without

Getting started: freelancer setup in 15 minutes

If you're new to amux, here is the fastest path from zero to agents running on real client work:

1. Install amux (2 minutes).

   git clone https://github.com/mixpeek/amux && cd amux && ./install.sh

   Requires Python 3.9+ and Claude Code CLI (or Codex CLI). That's it — no external services, no Docker.
2. Register your first client project (2 minutes).

   amux register client-a --dir ~/Dev/client-a-app --yolo
   amux serve  # → https://localhost:8822

3. Write a CLAUDE.md for the project (5 minutes). Add project context, coding conventions, and two or three "gotchas" you've learned about the codebase. This is the most important investment you'll make — better context means better overnight output. See the getting started guide for a template.
4. Queue one task and run one agent (3 minutes). Start with a well-scoped, isolated task — not your most critical feature. Watch the output. Review the PR. Get comfortable with the quality bar before running overnight multi-agent sessions.

   amux board add "Add a loading spinner to the data table while fetching" --session client-a
   amux start client-a

5. Add your clients to the CRM (2 minutes).

   amux crm add "Sarah Chen" company="Acme Corp" email="[email protected]" role="CTO"
   amux crm add "Marcus Roy" company="Beta Tech" email="[email protected]" role="VP Eng"

6. Install the iOS app so you can check in from anywhere. It connects to your local amux server via the same URL — the dashboard is a PWA, no App Store build required for the web version, but the native iOS app gives you push notifications when tasks complete. See the mobile management guide.

Most freelancers are running their second and third client sessions within the first week. The Concierge offers a hands-on setup session if you'd rather skip the configuration steps entirely.

Frequently asked questions

Can I realistically handle more clients using AI coding agents?

Yes. With amux, you run one dedicated agent per client project simultaneously — each isolated in its own git worktree, following that project's CLAUDE.md. The bottleneck shifts from coding hours to review speed. Freelancers report handling 3-5x their previous client load without hiring.

How does overnight delivery work?

Queue the client's feature requests before you stop for the day. Agents run through the night with YOLO mode enabled. In the morning you review the completed PRs, do a quick QA pass, and deliver. Clients get near-next-day turnaround without you working nights.

What does amux cost?

amux is free and MIT licensed. You pay only for AI model tokens. Claude Max at $200/month covers unlimited overnight runs. API pricing runs $10-30/night for a 5-agent session. At typical freelance billing rates, the ROI is clear after one client delivery. See the full cost breakdown.

How do I keep client projects isolated?

Register each client as a separate amux session with its own working directory and CLAUDE.md. Agents never touch files outside their registered directory. The board tracks tasks per session so client work never bleeds across projects.

Can I manage client relationships inside amux?

Yes. The built-in CRM stores contacts, logs interactions, tracks follow-ups, and integrates with the Gmail API for sending client emails — with your real signature, in the correct thread — all from the CLI or dashboard.

git clone https://github.com/mixpeek/amux && cd amux && ./install.sh
amux register client-a --dir ~/Dev/client-a --yolo
amux start client-a
amux serve  # → https://localhost:8822

Related resources: Getting Started Guide · Mobile Management PWA · AI Coding While You Sleep · Team Workflow Setup · Agent Cost Guide · Concierge