| Product | Firetower |
| GitHub | firetower-cloud/firetower |
| Category | Developer Tools / Agent Infrastructure |
| License | Open Source |
| Built With | Rust |
| Agents Supported | Claude Code, Codex, Kimi Code, and others |
What Is Firetower?
Firetower lets you run AI coding agents on your own servers instead of your laptop. You connect any machine — a dev server, a cloud VM, a spare desktop — and orchestrate Claude Code, Codex, or other coding agents through an SSH tunnel. Sessions survive disconnects, run in isolated Git worktrees, and can be accessed from your desktop or phone.
The motivation is practical. AI coding agents are CPU and memory hungry. They run best on powerful machines with fast SSDs and plenty of RAM. Your laptop may not be that machine. Firetower decouples the agent runtime from the machine you’re sitting at.
Key Features
Agent-Agnostic
Firetower supports Claude Code, Codex, Kimi Code, and can be extended to other coding agents. You pick the agent for the task, and Firetower handles the runtime. This matters if you’re evaluating agents or switching between them depending on the codebase.
Persistent Sessions Over SSH
Sessions run on your server and persist through network interruptions. Close your laptop, commute, open it again, and your agent session is where you left it. No lost context, no re-running prompts. The connection goes through an SSH tunnel, so your code never touches a third-party server.
Isolated Git Worktrees
Each agent session operates in its own Git worktree. Multiple agents can work on the same repository simultaneously without stepping on each other’s changes. When you’re ready, you review and merge each worktree independently.
Parallel Agents
Run multiple agents on the same server, each in their own isolated session. If you’re working on a feature branch and want to fix a bug on another branch at the same time, Firetower keeps both agents running and isolated from each other.
Mobile Access
Monitor and interact with your agent sessions from your phone. Start an agent on a task at your desk, then check on it from your phone while away. The mobile interface shows session status, allows basic interaction, and sends push notifications when an agent needs input.
Who Is This For?
- Developers with remote servers who want to offload agent compute from their laptop
- Teams running multiple agents in parallel on shared infrastructure
- Anyone who needs persistent agent sessions that survive disconnects
- Security-conscious teams that want agent sessions on infrastructure they control
Pros
- Open source and self-hostable
- Rust-based — fast and memory-efficient
- Agent-agnostic (Claude Code, Codex, Kimi Code)
- Sessions persist through disconnects
- Isolated Git worktrees for parallel agents
- Mobile access with push notifications
- SSH tunnel — no third-party servers
Cons
- Requires your own server infrastructure
- Setup complexity: SSH tunnels, server provisioning
- Young project (v0.41, rapid iteration)
- Rust dependency — building from source requires Rust toolchain
- No hosted option — self-host or nothing
Verdict
Firetower addresses the infrastructure gap in the AI coding agent ecosystem. The agents themselves are powerful, but running them on a laptop has real limitations: battery drain, fan noise, competing for resources with your other work. Firetower moves the compute to a server and gives you a clean interface to manage it.
The isolated Git worktree model is the right approach for parallel agents. Without isolation, multiple agents modifying the same repository create merge conflicts and race conditions. Firetower prevents this architecturally.
The main barrier is that you need server infrastructure. This isn’t a consumer product — it’s a power-user tool for developers who already have servers and know how to use SSH. For that audience, Firetower is a compelling addition to the AI development stack.
Build a developer tool?
Get reviewed and linked from our 27-site network. Placements live in 48 hours.
See Packages