Aetower
Docs
Aetower Feature List
Aetower is a local-first macOS observability and operator dashboard. It helps a technical user, developer, or AI agent understand what is making a Mac slow, hot, noisy, memory pressured, or expensive to run. This document lists the major features in both practical human terms and technical implementation terms.
Feature overview
| Area | Human-readable value | Technical capability |
|---|---|---|
| Live monitor | See what is hurting the Mac right now. | Groups running processes into entities, ranks them by friction, and shows host pressure signals. |
| Entity friction scoring | Replace dozens of raw counters with one triage number. | Combines CPU, memory, disk, network, wakeups, energy, thermal, and anomaly signals into a 0-100 score. |
| Entity detail and provenance | Understand what an item is, where it came from, and what to do next. | Shows process membership, launcher/origin context, trend data, recommendations, and action planning. |
| History | Reconstruct past slowdowns instead of guessing. | Persists local snapshots and supports historical investigation, comparison, and store-health checks. |
| Timeline | Read a machine incident as a story. | Correlates lifecycle events, host state changes, sensor alerts, restart loops, and AI session markers. |
| AI agent observability | Measure the local cost of coding agents and model runtimes. | Detects supported AI runtimes, attributes burden, estimates GPU/VRAM pressure, and tracks session energy/cost context. |
| Storage hygiene | See reclaimable storage without confusing logical size with local disk savings. | Reports logical bytes, APFS physical-block estimates, hardlink dedupe, sparse/cloud placeholder flags, purgeable capacity, and clone-lineage caveats. |
| Local MCP server | Let trusted local agents inspect Aetower data without a second collector. | Ships a local MCP interface over an app-owned socket and stdio helper; guarded operator actions are visible by default and can be hidden in Settings. |
| Diagnostics | Debug Aetower itself when collection, adapters, or persistence misbehave. | Exposes subsystem events, pipeline timing, capability state, session health, and support-bundle manifests. |
| Fleet | Compare nearby Macs without a cloud account. | Uses local peer discovery to surface summary machine health across Aetower peers. |
| Settings and setup | Tune depth versus overhead safely. | Controls collection cadence, optional integrations, MCP registration, export behavior, and reset/support flows. |
| Export and observability integrations | Send selected telemetry to an existing stack when needed. | Supports optional OTLP/HTTP metrics export and documented dashboard/collector workflows. |
| Privacy and safety controls | Keep local machine data understandable and deliberate. | Keeps collection local by default, documents observed data, and supports privacy-aware export/support workflows. |
1. Live Monitor
What it does for people
The Monitor view is the first place to look when the machine feels slow. Instead of forcing the user to scan hundreds of raw processes, Aetower groups related processes into understandable entities such as an app, daemon family, terminal session, local AI runtime, or browser group. The highest-cost entities appear at the top so the operator can quickly answer: "What should I look at first?"
Technical details
- Samples host and process-level runtime signals.
- Groups raw processes into higher-level entities using identity, origin, runtime, repository, workspace, and launcher context where available.
- Surfaces host-level health such as CPU pressure, memory pressure, wakeups, thermal state, battery context, and active entity count.
- Presents ranked entities and drill-down detail rather than only process IDs.
2. Friction scoring
What it does for people
Friction is Aetower's shorthand for "how much this thing is costing the machine right now." A high friction score does not simply mean high CPU. It can also mean memory pressure, wakeups, disk churn, network activity, energy impact, thermal contribution, or an unusual change from recent behavior.
Technical details
- Produces a 0-100 score for triage and ranking.
- Combines multiple burden dimensions instead of relying on one counter.
- Separates host pressure from per-entity contribution so the user can see both the overall machine condition and likely causes.
- Feeds recommendations, anomaly explanations, trend cards, and timeline events.
3. Entity detail, provenance, and action planning
What it does for people
When an entity looks suspicious, the detail view explains what it is, which processes belong to it, why it is ranked highly, and what actions are safe or reasonable. This is intended to reduce the "is this process important?" anxiety that often happens in Activity Monitor.
Technical details
- Shows entity process membership and tree relationships.
- Uses process origin metadata to explain launch paths and runtime context.
- Provides recommendations based on dominant burden signals and environment context such as battery or thermal pressure.
- Supports action planning through controller surfaces rather than blindly killing processes.
4. History
What it does for people
History lets the user investigate what happened earlier. If the Mac was sluggish during a meeting, the user can go back to that time window and inspect which entities were active, which scores were high, and whether host pressure changed.
Technical details
- Persists snapshots locally for post-incident review.
- Supports paged historical data access and before/after comparison workflows.
- Tracks history store health, quality, gaps, ordering, and retention-related metadata.
- Enables investigations that do not require cloud telemetry.
5. Timeline
What it does for people
Timeline turns raw monitoring data into a chronological narrative. It helps the operator see the order of events: a process spiked, memory pressure rose, thermal state changed, an AI session ended, or a restart loop appeared.
Technical details
- Emits entity lifecycle and behavior events.
- Correlates host state changes with entity changes.
- Includes sensor and pressure alerts when available.
- Records AI-session and runtime markers for later correlation.
6. AI agent and local model observability
What it does for people
Aetower makes the cost of local AI work visible. It helps answer whether a coding agent, local model server, transcription tool, or inference runtime is consuming excess CPU, memory, GPU-like resources, battery, or project budget.
Technical details
- Detects and groups common local AI tools and model runtimes.
- Surfaces runtime burden leaders, delegated-session context, approval queue context, and recent AI-related changes where integrations provide them.
- Estimates GPU attribution and VRAM/unified-memory pressure when direct per-process GPU data is unavailable.
- Integrates optional Chau7 context for AI session state, project cost, and adapter metadata.
7. Storage hygiene and APFS-aware estimates
What it does for people
Storage hygiene helps the operator find rebuildable or redundant local data without overstating how much space cleanup will actually free. On APFS, logical file size can differ from local allocated blocks because of sparse files, compression, cloud placeholders, hardlinks, and cloned/shared extents.
Technical details
- Tracks logical bytes and local physical-block estimates separately.
- Uses local allocated blocks for reclaim estimates when available, and treats zero-block cloud or sparse placeholders as 0 bytes of proven local reclaim.
- Deduplicates hardlinked files within a sized directory while warning that external hardlinks can still reduce the space actually freed.
- Reports volume free-now, available, important/opportunistic available, and purgeable-capacity estimates where the platform exposes them.
- Labels sparse/shared-block candidates as estimates; Aetower does not infer or promise exact APFS clone lineage.
8. Local MCP server
What it does for people
Trusted local AI agents can ask Aetower what is happening on the machine. This means an agent can inspect machine pressure, entity details, history, diagnostics, and recommendations without launching its own duplicate monitoring engine.
Technical details
- Runs an app-owned local MCP server when Aetower is active.
- Uses a local Unix socket and packaged stdio helper for supported clients.
- Offers one-click registration for supported Claude and Codex clients; automatic registration remains off by default.
- Exposes tools for snapshots, host summaries, entity details, diagnostics, history pages, recommendations, support-bundle manifests, runtime lag, export queries, investigation bundles, and guarded operator actions. Operator actions are visible by default, can be hidden in Settings, and remain preview- and approval-gated.
- Standard cached tools can still provide last-known state when the app is not running; deeper profiling requires the live app.
9. Diagnostics and self-observability
What it does for people
Diagnostics explains whether Aetower itself is healthy. It is useful before filing an issue, sharing a support bundle, or investigating why a sensor, adapter, database, or collection path is unavailable.
Technical details
- Tracks subsystem events for engine, collector, adapters, persistence, history, runtime, GPU-related sampling, and export paths.
- Surfaces pipeline timing so expensive collection phases can be spotted.
- Reports capability status and permission/adaptor availability.
- Provides support-bundle previews and diagnostics summaries for issue reports.
10. Fleet
What it does for people
Fleet gives a lightweight view of nearby Aetower machines with Fleet enabled, useful for comparing a MacBook, Mac Studio, build host, or teammate machine on the same trusted network. It is meant for local awareness, not enterprise cloud monitoring.
Technical details
- Uses opt-in local network peer discovery.
- Displays summary host health such as CPU, memory pressure, thermal state, and active entity count.
- Avoids requiring a centralized account or external service for basic peer awareness.
11. Settings, setup, and runtime tuning
What it does for people
Settings lets users choose how much detail Aetower collects and which optional features are enabled. The default path is designed for Developer Preview users who want useful observability without turning on every advanced capability.
Technical details
- Controls UI refresh, engine collection cadence, GPU sample cadence, and full-collection behavior.
- Configures optional integrations such as Chau7, Chromium-compatible debug endpoints, Docker, telemetry export, and advanced helper paths.
- Manages local MCP client registration.
- Includes setup, diagnostics, support, and reset workflows.
12. Export and external observability
What it does for people
Some teams already use OpenTelemetry, Prometheus, Grafana, or other internal observability systems. Aetower can optionally export selected metrics so local Mac performance can be investigated alongside the rest of a developer platform.
Technical details
- Supports optional OTLP/HTTP metric export.
- Documents collector and dashboard setup for local or enterprise use.
- Keeps export opt-in so users do not accidentally send local machine metadata outside their device.
13. Privacy, safety, and Developer Preview boundaries
What it does for people
Aetower observes sensitive local machine metadata, so the product needs clear privacy expectations. The default behavior is local-first, and users should know which optional features increase visibility or require extra permissions.
Technical details
- Local collection is the default operating model.
- Optional exports and integrations must be configured deliberately.
- Developer Preview builds avoid assuming production readiness.
- Documentation calls out observed data classes, known limitations, signed build expectations, and reset/support guidance.
Feature maturity notes
Aetower is currently a Developer Preview. The core monitor, history, diagnostics, settings, and local MCP workflows are documented as primary product surfaces. Some integrations depend on macOS permissions, local services, supported AI clients, adapter availability, or build configuration. Treat optional helpers, export paths, and advanced integrations as deliberately configurable rather than always-on behavior.
Aetower is a free early-alpha download for macOS 14+ (Apple silicon).
Download for macOS