How-tos

Short, task-based walkthroughs. Follow one when you need it. New to Aarmos? Start with the CLI quickstart.

See it work

All walkthroughs

Add your LLM provider

BYO-key or fully local — Ollama, Google, Groq, OpenAI, OpenRouter, or any OpenAI-compatible endpoint you configure yourself.

Connect your first connector

Turn on a built-in tool from the Connectors catalog.

Wire a custom REST API

Base URL + auth + endpoints (or import OpenAPI 3).

Build an agent

Install a starter or author your own with scoped tools.

Lock the vault

Device password, auto-lock, recovery key.

Move to a new device

Export and restore an encrypted backup.

Manage storage & retention

Per-store usage, retention windows, on-demand purge.

Restore governance from a Recovery Point

Verify offline, close every named gap, then resume — all-or-nothing.

Set guardrails: rate limits, spend cap, panic stop

Per-tool caps, USD budget, kill-switch shortcut.

Observe a local runtime

Pair the Operations view with your machine using a short-lived code.

Wire up the CLI bridge

Pair the PWA to a localhost CLI with dry-run defaults.

Run a playbook headless

Simulate, sign, run, and attest — with a kill-switch file.

Framework compatibility matrix

Which agent frameworks are Verified, Evaluation, or Planned — with links to runnable examples.

Govern a LangChain agent (60s)

One env var routes an unchanged LangChain.js agent through the local gate.

Govern a CrewAI crew (60s)

One eval line routes a stock CrewAI crew through the authenticated local gate.

Govern an AutoGen agent (60s)

Same env-var pattern for AutoGen (agentchat) single-agent and team runs.

Several controls overlap. Reach for the narrowest one that solves the problem.

Which control to use in which situation
Situation	Use	Where
One agent misbehaves; the rest are fine	Revoke or pause that agent	Registry
Active incident; stop activity now	Kill switch (all, writes, or destructive)	Kill switch
Approval needed above an input threshold	Policy threshold	Settings → Policy thresholds
A tool should be permitted or refused by default	Permission scope	Permissions
Cap call volume or spend	Rate limits and spend ceiling	Set guardrails

Settings changes are recorded in the local ledger and can be exported.

Need something we didn't cover? Ask us.