Launch
Start supervised agent processes remotely using approved launch profiles, runtime-aware options, and safe workspace selection.
Vendor-neutral control plane
Mobile control plane for coding agents.
Launch, monitor, and intervene in terminal-native coding agents running on your own machines, with Gemini CLI as the flagship integration and a vendor-neutral adapter model underneath.
Gemini-first depth
Vendor-neutral adapters
Built for intervention, not just chat
Fleet Snapshot
Online Machines
1
Running Agents
3
Warnings / Stuck
1
Queued Tasks
2
running
Gemini Local CLI
Summarizing repo changes on workstation-main
warning
Codex Local CLI
No output for 82s, surfaced for review
queued
Hermes via WSL
Waiting for free worker capacity
Product at a Glance
A cleaner operator surface for launch, monitoring, and intervention.
Mobile Agent Control exists because coding agents are no longer a single-terminal activity. Once work spreads across machines and runtimes, you need a control layer that stays fast, honest, and operational.
Dashboard screenshot placeholder
Operator Home
Fleet Summary
Running Now
Machine Health
Recent Activity
Launch
Start a new agent on the right machine with a safe workspace and launch profile.
Monitor
Watch state, heartbeat, last output, and machine health in one place.
Intervene
Restart, stop, retry, or prompt agents as soon as the situation changes.
Monitor
Track active runs through fleet summary, running agents, machine health, logs, and recent activity instead of polling terminals manually.
Intervene
Stop, restart, retry, and send prompts quickly when long-running work stalls, fails, or needs steering.
Why this exists
Vendor CLIs are strong local tools, but they do not give you a unified mobile control surface for multiple machines, long-running jobs, and runtime-neutral operations.
What It Is
A remote control and monitoring layer for coding agents running on your own machines.
Mobile Agent Control sits above machine-local agent runtimes and gives you an operator-grade surface for launch, monitoring, lifecycle control, logs, machine health, and recent activity.
The Android client and responsive web console are both thin control surfaces. The machine-side FastAPI supervisor owns lifecycle state, worker capacity, websocket event streaming, runtime adapter integration, and audit history.
Control Plane Positioning
- Remote launch and control for terminal-native agent runtimes
- Vendor-neutral machine, agent, task, and audit model
- Gemini-first depth without collapsing the architecture into a single vendor
Why It Exists
Coding agents do not stay neatly bound to one terminal on one machine.
Multiple machines
Desktop, laptop, and remote boxes drift apart quickly when agents are running in different places.
Long-running work
Agent runs can last long enough that you need to check status, logs, and health without staying at the desk.
Fast intervention
When something stalls or fails, you need stop, restart, retry, and prompt actions immediately, not after reconnecting manually.
Neutral control layer
Vendor-specific CLIs are strong local tools, but they are not a unified remote fleet operations surface.
Key Capabilities
High-signal operator workflows built around real agent lifecycle control.
Launch agents remotely
Start new agent processes on selected machines using safe launch profiles, approved workspaces, and runtime-aware options.
Monitor running agents live
See state transitions, elapsed runtime, heartbeat, last output, logs, and warning or stuck indicators in near real time.
Intervene fast
Stop, restart, retry, and prompt active agents from mobile or web without dropping into a terminal session first.
View machine health and activity
Track worker capacity, queue depth, resource hints, recent launches, completions, failures, and supervisor trust signals.
Gemini-first, adapter-driven
Gemini CLI is the flagship integration today, but the architecture is intentionally runtime-neutral through CLI adapters and shared supervisor models.
Dashboard Spotlight
An operator console home screen, not a pile of admin panels.
Dashboard screenshot placeholder
Total Connected Machines
1
Running Agents
3
Warnings / Stuck
1
Queued Tasks
2
Running Now
running
Gemini Local CLI
workspace: project-repo
warning
Codex Local CLI
last output 82s ago
Machine Health
Heartbeat: live
Capacity: 1 / 2 busy
CPU: if available
Memory: if available
Recent Activity
Launches, completions, failures, restarts, stop requests, and machine transitions appear here.
One surface for remote operations
The dashboard is the first screen because it should answer the operational questions immediately: what is healthy, what is running, what is stalled, and where should the next launch go.
Fleet summary
Understand total machine state, online vs offline reachability, running agents, queue pressure, and recent completions in a glance.
Running agents
See runtime type, machine, workspace, elapsed time, last output, and fast actions where intervention matters most.
Machine health
Use supervisor heartbeat, capacity, and optional CPU or memory visibility to decide where work should go next.
Recent activity and quick actions
Keep launches, failures, restarts, and retries close to the top-level operator experience instead of buried in detail screens.
How It Works
Control UI on top, machine supervisor in the middle, runtime adapters underneath.
Phone / Browser
Android + Responsive Web
→
Control UI
Dashboard, launch, monitoring, intervention
→
Machine-side supervisor
FastAPI orchestration + lifecycle state
Runtime adapters
Gemini CLI, Codex CLI, Hermes, more
Coding agent CLIs
Real local processes, logs, prompts, restarts
Pre-release transport
Private connectivity over Tailscale
Future control plane
Cloud-routed transport without UI rewrite
Why This Is Different
Built as a control plane, not a wrapper around a single terminal workflow.
Not tied to one vendor
Shared machine, agent, task, audit, and launch-profile models stay neutral while adapters handle runtime-specific behavior.
Not just local terminal UX
The machine supervisor owns lifecycle state, worker capacity, logs, websocket events, and operator-visible audit history.
Not just chat on mobile
The product is centered on launch, monitoring, health, intervention, and routing decisions across agent workloads.
Gemini-first wedge
Gemini CLI is the deepest supported runtime today, but the product is positioned to extend across other terminal-native coding agents cleanly.
Roadmap
Practical evolution from strong MVP to fuller control plane.
Current
Working MVP
Android and web operator surfaces, supervisor lifecycle control, monitoring, adapters, workspaces, MCP visibility, and Gemini-first runtime support.
Next
Monitoring hardening
Stale-data trust signals, reconnect handling, stronger machine health heuristics, and better failure surfacing.
Next
Dashboard polish
Sharper prioritization, richer recent activity, screenshots, and tighter fast-intervention workflows.
Reliability
Persistence and recovery
Stronger state durability, recovery semantics, and operational trust under supervisor restarts or failures.
Surface
Browser-first control plane
Continue strengthening the responsive web console so it stands as a first-class operator entry point.
Future
Cloud-routed architecture
Move from private direct reachability to a brokered control plane without rewriting the product model.
Technical Details
Deeper technical context stays close, but below the main product story.
Architecture
Read the current control-plane layout, runtime adapter model, and transport abstraction.
Open architecture docAPI and setup
Use the repo README for endpoint coverage, launch profiles, local setup, and runtime notes.
Open setup guideSource and issues
Browse the code, open issues, or track the latest changes in the public repository.
Open repositoryImplementation status
Current builds persist recent supervisor state in SQLite while live scheduling and coordination remain in memory, with recovery hardening still on the roadmap.