Open source, local-first
Your PC keeps working. You answer the ask.
AgentMob turns Codex, Claude Code and Antigravity into a conversation on your phone. Approve a risky step from anywhere; your history stays in SQLite on your PC.
v1 is in development. The npm package is not published yet.
- 3 agents, one inbox
- 0 bytes we store
- SQLite on your own disk
Work continues while the phone is in your pocket.
Three screens, no dashboard. The PC is the source of truth, the phone is the remote.
-
It works on your PC
Codex, Claude Code and Antigravity share the same session list.
-
It asks before it acts
A risky step waits for a tap. Nothing runs behind your back.
-
It keeps going
Answer once, scope it to the session, or deny. Then close the app.
What it actually looks like.
Rendered from the real components. No stock screenshots, no invented dashboards.
-
Sessions
Three agents, one list, no vendor tab to remember.
-
Permission sheet
The exact input, the risk, and three scoped answers.
-
Audit trail
What was approved, from which device, and when.
-
Voice note
Dictation goes to Groq, then is deleted. The screen says so.
-
Pairing
A QR code and a short code. No account, ever.
Limits you should know
Written down before you install, not after.
-
One phone in v1
Multi-device pairing is planned for v1.1. Losing the phone means pairing again.
-
No app-layer encryption yet
WebRTC uses DTLS. A libsodium or Noise layer is required in v1.1, not in v1.
-
Agent support is not uniform
Codex targets full support. Claude Code and Antigravity start at basic and earn full after their contract tests.
-
Desktop only
Windows, macOS and Linux. No agent runs on the phone.
-
No offline queue
If the PC sleeps, nothing is answered for you.
-
A compromised PC is compromised
The server runs as you, with your rights.
Start with the machine you already use.
Install the local server, pair one phone, and keep the work on your PC.