I Built a Tool to Run Claude Code From My Phone. It Became a Product.
The Problem I Was Solving for Myself
I run Claude Code constantly. It handles client work, internal tooling, marketing copy, code reviews — essentially anything that can be delegated to a capable AI running in a terminal. The problem was that I could only use it when I was sitting at my desk.
I wanted to kick off a long-running task from my phone and check on it later. I wanted to be away from my desk and still spin up a session. I wanted to see what Claude had accomplished without having to SSH in blind and dig through files.
So I built something.
The First Version: A Phone-Controlled SSH Dashboard
The first iteration was simple: a web dashboard running on my local machine, exposed via Tailscale (a zero-config VPN), with the Terminus iOS app as the mobile SSH client. I could navigate to a URL on my phone, see my running sessions, and control them.
The first time it worked — I mean actually worked, not "technically running" — I was at a coffee shop. I fired off a message in the dashboard, watched the session pick up the task, and checked the output 20 minutes later on my walk home.
It sounds small. It wasn't. That feedback loop changed how I used AI.
What Grew Out of It
Once I had the basic working model, the gaps became obvious fast:
- No persistence. If the process died, the session was gone — no logs, no history, no way to audit what happened.
- No team access. I couldn't hand a session to someone else or let a client observe a working session.
- No billing controls. Everything was free and local, which worked for me but not for a product.
- No Windows support. The agent relied on Unix PTY primitives. Getting it working on Windows — where most of the potential users are — required a complete rewrite of the spawn logic.
So I rewrote it. The Go-based agent that runs today handles Windows via ConPTY (Microsoft's pseudoconsole API), falls back through Git Bash and WSL tiers, and only drops to raw pipes as a last resort. Session recording is stored as gzip-compressed NDJSON on Google Cloud Storage. The web dashboard is a full Next.js app with Stripe billing, Google OAuth, GitHub OAuth, team RBAC, webhooks, and API keys.
The Market Timing Wasn't Planned
I didn't set out to displace Teleport. I set out to scratch my own itch.
But while I was building, three things happened in the market:
1. Teleport killed their free tier in June 2024 — users who had been relying on it for SSH access management suddenly needed a paid alternative starting at $70/seat/month. 2. Cloudflare deprecated their SSH audit product (Cloudflare Access SSH session recording) effective March 16, 2026. 3. ngrok started throttling free plans, making it significantly less useful for the lightweight SSH tunneling use cases it had dominated.
The teams who got caught by these changes needed something: SSH access management, session recording, audit logs, team access controls. Not $70/seat. Not an enterprise contract. Something they could start free and scale into.
That's exactly what SessionForge is.
What It Is Now
SessionForge is a session management platform for teams running AI agents and developer tools remotely. The core loop is:
1. Install the Go agent on any machine (Windows, Mac, Linux — one command) 2. The machine registers with the SessionForge cloud, gets a machine ID 3. From the web dashboard or API, spin up a session on any registered machine 4. Watch it run in real time, record it, share it, audit it
Free plan gets you one machine, three sessions, one day of history. Pro is $19/month for five machines, unlimited sessions, 30 days of history, session recording, webhooks, and API access. Team and Enterprise add shared seats, RBAC, IP allowlisting, and SSO.
Why I'm Writing This
I write this because the origin matters. SessionForge wasn't built by a team with a market thesis and a Series A. It was built by one person who needed the thing, built it for himself, and then watched the market create the exact gap the product fills.
That's not a disadvantage. That means every feature in it was built because someone — me — actually needed it to work.
If you're running Claude Code for client work, managing AI sessions across machines, or looking for a lightweight alternative to Teleport, try SessionForge free. No credit card required.
Related Articles
Why Your AI Needs a Self-Improvement System (And What One Actually Looks Like)
AI tools are only as good as what they remember. If your system forgets every correction the moment a session ends, you're not building anything — you're just repeating yourself. Here's why self-improvement matters, what it can and can't do, and how we built one that runs every night.
We Built an AI That Learns From Its Own Mistakes — Every Night
Most AI tools forget everything the moment a session ends. We built a self-improvement pipeline that automatically captures corrections, validates them, and deploys better behavior every night at 2 AM. Here's how it works and what 4 months of data shows.
What Is Claude Code and Why Should Your Business Care?
You've probably heard of ChatGPT. Claude Code is different — and for business owners who aren't developers, it's worth understanding what it actually does and how it gets deployed to save time and money.