First Run

First Health Check

Open the desktop app and check the Runtime / Workbench status first. It should show:

• Core is installed or ready to start.
• Core is reachable after start.
• The owner token is present.
• The profile, root, runtime paths, and ports are consistent.
• Local Gateway can advertise explicit tool capabilities.
• Provider planning is configured when you expect AI tool-use.

Advanced operators can still run CLI diagnostics against the same Core:

aegis status
aegis onboarding doctor
aegis diagnostics deployment-readiness
aegis worker tools --no-exec

Configure Provider Planning

Aegis Manager uses a real AI provider for natural-language planning. Set provider values in the desktop app settings, the managed Core profile, or an operator-controlled secret store:

AEGIS_LLM_BASE_URL=https://api.example.com
AEGIS_LLM_MODEL=your-model
AEGIS_LLM_API_KEY=...

Restart or refresh the profile after changing provider configuration.

No keyword shortcuts

Aegis should not infer weather, reminders, terminal commands, web search, or scripts with keyword matching. Natural-language requests go through Manager planning, then become structured tool or task calls.

What to Try First

Ask a simple status question

Use Chat first to confirm that Aegis can answer through the current Core and provider path. Use CLI checks only for operator diagnostics.

Run a readiness check

Verify worker tools without executing them with aegis worker tools --no-exec.

Open Workbench

Inspect Core health, tasks, Native Nodes, and evidence from the desktop app when available.

Pair mobile later

Pair mobile devices only after Core URL, token, and network path are stable.

Safe Operating Model

1. You ask for something in Chat, a mobile app, an agent plugin, or a channel.
2. Aegis Manager plans with a real provider.
3. The plan emits structured tool or task calls.
4. Workers, Local Gateway, or Native Nodes execute explicit payloads.
5. Evidence is stored and linked back to the result.

High-risk actions, such as deleting files, changing system settings, controlling GUI, sending external messages, or publishing content, should go through explicit approval.