AnySee

AnySee is a customizable macOS menu bar attention center for signals you actually want to watch. It is not a notification center, not a full dashboard, and not a feed. The first version is local-first: configuration, scripts, credentials, and state stay on the Mac by default.

Domain: anysee.bar

MVP Scope

• Native macOS app built with Swift 6, Swift Package Manager, SwiftUI, AppKit, and NSStatusItem.
• Menu bar status icon reflects the overall attention state.
• Popover shows an Attention Feed grouped by priority and state.
• Local config directory:

~/Library/Application Support/AnySee/
  anysee.toml
  sources/*.toml
  scripts/*
  schemas/signal.schema.json
  examples/*
  AGENTS.md

• Source types:
  • manual: static signal entries in TOML.
  • http: request a URL and produce a signal on status mismatch or JSON field condition.
  • script: run a local executable and parse stdout as Signal JSON.

Not included in the MVP: Gmail, Slack, GitHub integrations, OAuth, cloud sync, AnySee backend services, plugin marketplace, or built-in AI API calls.

Build

swift build
swift test

Run the menu bar app from source:

swift run AnySee

Run the CLI:

swift run anysee doctor
swift run anysee validate
swift run anysee preview welcome
swift run anysee run welcome

doctor creates the local config directory and default examples if they do not already exist.

Website

The first-launch website for anysee.bar lives in public/. It is a static Cloudflare Pages site with one optional Pages Function at functions/api/waitlist.js for waitlist/contact capture.

Cloudflare Pages settings:

Build command: none
Build output directory: public
Functions directory: functions
Project name: anysee-bar

Local preview with Wrangler:

npx wrangler pages dev public --kv=ANYSEE_WAITLIST

First deploy:

npx wrangler pages deploy public --project-name anysee-bar

The waitlist Function stores submissions only when a Cloudflare KV binding named ANYSEE_WAITLIST is configured. Do not commit API tokens, mailbox credentials, or webhook secrets. If you add outbound email later, store the secret in Cloudflare environment variables and keep only the non-secret binding or variable name in the repo.

Configuration

Global settings live in anysee.toml:

app_name = "AnySee"
quiet_by_default = true
max_signals_in_popover = 30

Each file under sources/*.toml defines one source:

id = "welcome"
name = "Welcome"
kind = "manual"
enabled = true

[refresh]
kind = "manual"

[[signal]]
id = "welcome-to-anysee"
title = "AnySee is ready"
body = "Edit sources/*.toml to decide what deserves attention."
priority = "normal"
state = "needs_attention"
source = "welcome"
url = "https://anysee.bar"

Signal priorities are none, low, normal, high, and critical. Signal states are ok, needs_attention, running, paused, and unknown.

Configure With AI

The app has a “Configure with AI” action that copies a prompt to the clipboard. It does not call GPT, Claude, Codex, or any model API. Paste the prompt into the AI tool of your choice and ask it to generate or modify files under the local config directory.

The generated prompt includes:

• AnySee product rules.
• Current config directory path.
• Existing source summary.
• Supported source types.
• Signal JSON example.
• Validation commands.
• Security constraints.

The config directory also contains AGENTS.md, written specifically for AI coding agents.

Signal JSON

Script sources must write either one Signal JSON object or an array of Signal JSON objects to stdout:

{
  "id": "backup-failed",
  "title": "Backup failed",
  "body": "Last backup exited with code 1",
  "priority": "high",
  "state": "needs_attention",
  "source": "backup-script",
  "url": "file:///Users/me/backup.log",
  "actions": [
    { "label": "Open log", "type": "open_url", "url": "file:///Users/me/backup.log" }
  ]
}

Supported actions are open_url, copy_text, run_command, dismiss, and snooze. run_command is executed directly without a shell. Absolute commands run as provided; relative commands resolve under ~/Library/Application Support/AnySee/scripts/ and must stay inside that directory.

Credential Policy

Do not put credentials in TOML, scripts, examples, prompts, or signal bodies. Future token-based sources should store secrets in macOS Keychain and reference only a non-secret key name from configuration.