feat: wrap AI, Images, Analytics Engine, Vectorize and RateLimit Bindings to accept native Python objects

[ryanking13]

Now that we have a good BindingWrapper class, these just work. Unfortunately AI and Vectorize bindings require a real prod account even for the local testing, so we cannot run unittest in CI.

I verified them working my making a demo worker with my local account + patched sdk.

Vectorize test worker

This requires 64-dimension DB in the account

import traceback

from workers import Response, WorkerEntrypoint


class Default(WorkerEntrypoint):
    async def fetch(self, request):
        from urllib.parse import urlparse

        path = urlparse(request.url).path

        if path == "/probe":
            vec = self.env.VECTORIZE
            return Response.json({
                "type": str(type(vec)),
                "wrapped": hasattr(vec, "_binding"),
                "has_insert": hasattr(vec, "insert"),
                "has_query": hasattr(vec, "query"),
            })

        if path == "/describe":
            try:
                result = await self.env.VECTORIZE.describe()
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/insert":
            try:
                vectors = [
                    {"id": "vec-1", "values": [0.1] * 64},
                    {"id": "vec-2", "values": [0.4] * 64},
                ]
                result = await self.env.VECTORIZE.insert(vectors)
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/query":
            try:
                result = await self.env.VECTORIZE.query(
                    [0.1] * 64,
                    {"topK": 5},
                )
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/upsert":
            try:
                vectors = [
                    {"id": "vec-3", "values": [0.7] * 64, "metadata": {"label": "test"}},
                ]
                result = await self.env.VECTORIZE.upsert(vectors)
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/get":
            try:
                result = await self.env.VECTORIZE.getByIds(["vec-1", "vec-2"])
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/delete":
            try:
                result = await self.env.VECTORIZE.deleteByIds(["vec-1"])
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        return Response.json({
            "endpoints": ["/probe", "/describe", "/insert", "/query", "/upsert", "/get", "/delete"],
        })

AI test worker

import json
import traceback

from workers import Response, WorkerEntrypoint


class Default(WorkerEntrypoint):
    async def fetch(self, request):
        from urllib.parse import urlparse

        path = urlparse(request.url).path

        if path == "/probe":
            ai = self.env.AI
            return Response.json({
                "type": str(type(ai)),
                "wrapped": hasattr(ai, "_binding"),
                "has_run": hasattr(ai, "run"),
            })

        if path == "/run":
            try:
                result = await self.env.AI.run(
                    "@cf/meta/llama-3.3-70b-instruct-fp8-fast",
                    {"messages": [{"role": "user", "content": "Say hello in one word"}]},
                )
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                    "traceback": traceback.format_exc(),
                }, status=500)

        if path == "/models":
            try:
                result = await self.env.AI.models()
                return Response.json({"success": True, "result": result})
            except Exception as e:
                return Response.json({
                    "success": False,
                    "error": str(e),
                    "type": type(e).__name__,
                }, status=500)

        return Response.json({
            "endpoints": ["/probe", "/run", "/models"],
        })

[ask-bonk]

The change is straightforward. Let me review the complete picture:

1. _workers.py: Adds "Ai" and "VectorizeIndexImpl" to the _BINDING_TYPES set. This means when _EnvWrapper.__getattr__ encounters these JS binding types, they'll be wrapped with _BindingWrapper, which auto-converts Python objects to/from JS via python_to_rpc/python_from_rpc.

2. wrangler.jsonc: Adds a vectorize binding config for tests.

3. test_vectorize.py: A minimal test that just checks the binding is wrapped. Makes sense given vectorize needs a real remote database.

The change is minimal and clean. The only thing I notice is the absence of a parallel test for the AI binding — but the same rationale applies (AI needs a real account), and arguably if you can't even configure a local AI binding in wrangler, there's nothing to test. The PR description explains this clearly.

The code is correct — "Ai" and "VectorizeIndexImpl" are the JS constructor names used by workerd for these bindings, and adding them to the set is the right mechanism to get _BindingWrapper applied.

LGTM

github run