Gatesi examples
==================
Every Python example runs in TWO ways with no changes:

1. OFFLINE (default): no Worker, no key, no Telegram. It uses AgentGate.local(...), which applies the same rules as
   the real server, and approves automatically so you can see the whole flow in a few seconds.
2. LIVE: set two environment variables first and the very same file talks to your real Gatesi:
     AGENTGATE_URL = https://agentgate.YOURNAME.workers.dev
     AGENTGATE_KEY = agk_...   (Keys tab in the viewer)
   Approvals then arrive on your phone.

In Colab, one cell per example:

    import urllib.request, os
    BASE = "https://agentgate.YOURNAME.workers.dev"
    def get(name):
        req = urllib.request.Request(BASE + "/" + name, headers={"User-Agent": "Mozilla/5.0"})
        open(name.split("/")[-1], "wb").write(urllib.request.urlopen(req).read())
    get("sdk/agentgate.py"); get("examples/_common.py"); get("examples/local_test.py")
    # for LIVE mode, also: os.environ["AGENTGATE_URL"] = BASE; os.environ["AGENTGATE_KEY"] = "agk_..."
    %run local_test.py

Files
  local_test.py        the simplest tour: allow, deny, approve, budget (offline)
  python_function.py   protect your own functions with @gate.tool
  openai_loop.py       an OpenAI-style tool-calling loop using gate.dispatcher (model is simulated)
  anthropic_loop.py    an Anthropic-style tool_use loop using gate.dispatcher (model is simulated)
  langchain_tool.py    @tool over @gate.tool (needs langchain-core; skips itself if it is not installed)
  js_example.mjs       the same ideas in JavaScript (needs agentgate.mjs next to it; Node 18+)
  web_request.sh       plain curl, any language can do this