Mobile network priority
An active authenticated phone receives the highest connection priority across Omniscio-managed network activity. It is always on and has no setting to configure, and it applies to both the normal off-loop mobile transport and its in-process fallback.
On this page
What it is
Yes. An active authenticated phone now receives the highest connection priority across Omniscio-managed network activity.
This is always on; there is no setting to configure. It applies to both the normal off-loop mobile transport and its in-process fallback.
What it does not do
This is cooperative admission inside Omniscio, not operating-system packet scheduling. It does not cancel a transfer that already started, reorder packets already on the network, control a third-party webview, or throttle arbitrary sockets opened by an external child process. The one exception is the local agents Omniscio itself starts: their AI calls go through Omniscio's upload relay (see "Local agents' uploads" below).
Those boundaries are deliberate: mobile gets first access to capacity Omniscio controls without risking partial uploads, corrupting work, or starving background jobs indefinitely.
Where to find it
Nothing to open or configure — network priority is always on and has no setting. It applies wherever an authenticated phone talks to Omniscio, and the neighbouring systems it complements are described under Related.
How it behaves
How it works
Omniscio uses three admission lanes for managed network work:
- Mobile-critical — phone connection activity and phone-triggered actions start immediately.
- Interactive — ordinary user-facing work may yield for at most 150 milliseconds while a phone is active.
- Background — lower-urgency work may yield for at most 2 seconds while a phone is active.
A phone keeps priority for 15 seconds after authenticated activity. If it disconnects, Omniscio keeps a 10-second reconnect grace window so background work does not rush into the recovery gap. An idle browser tab cannot hold priority forever, and a desktop browser cannot claim the mobile lane.
What it protects
- Web Access requests made by an active phone.
- Omniscio HTTP work that uses the shared managed client.
- Background AI-provider calls that Omniscio explicitly marks as background.
- Both Web Access server transports, so fallback behavior matches the default path.
- Your upload line while cloud agents work. Cloud agents send their AI traffic through this computer and normally use 40–80% of your measured upload, letting your other traffic go first. While your phone is active they give up another tenth of the line (30–70%), and if you set a phone limit for the relay they never go above it.
Local agents' uploads
Every AI call a local agent makes re-sends its whole conversation, so a few dozen busy agents can fill your upload line on their own and leave nothing for your phone. New local Claude Code agents now send their AI calls through Omniscio's upload relay on this computer instead of straight to the AI company:
- Smaller uploads. Calls to Anthropic are compressed before they leave (a typical request shrinks by more than half).
- Your phone keeps its share. While Omniscio can measure your line, local agents take turns inside their own budget instead of fighting at your router: they get 75% of the line, less what your other Omniscio computers on the same Wi-Fi are sending, and a quarter always stays free for everything else in the house. When the line cannot be measured (for example on a Mac, or just after Omniscio starts) they run at full speed.
- Your own shares go first. While a recording or other share of yours is uploading, local agents slow to the same low floor as cloud agents, so the upload is not starved; they get their normal share back the moment it finishes.
- Nothing changes about sign-in. Each agent keeps its own Claude sign-in or key; Omniscio only passes it along to the same company the agent already uses, and stores none of it.
- It steps aside when it should. An agent starts exactly as before when the relay is not ready, when your computer uses an internet proxy, when the agent's own Claude settings set their own address or headers, or for engines other than Claude Code. Agents already running are never moved.
It is on by default. To turn it off, set localAgentRelayEnabled to false (ask your agent to
change it), or start Omniscio with AMC_DISABLE_LOCAL_AGENT_RELAY=1. If the relay itself stops,
routed agents pause their AI calls until it restarts; newly started agents go direct meanwhile.
Related
- mobile-remote-access.md — the phone's remote-session connection this priority order protects.
- main-heartbeat.md — the heartbeat that keeps a healthy phone connected.
Last verified 2026-11-08