feat(arize): add Browserbase × Arize AX integration

Use Browserbase as a load_page tool for a tool-calling agent, traced end to
end with Arize AX via OpenTelemetry + OpenInference. The model calls the
Browserbase tool to browse a page and answer; the run is exported to AX as a
CHAIN → LLM → TOOL trace with tokens and cost.

- src/agent.ts: tool-calling agent with manual CHAIN/TOOL spans
- src/instrumentation.ts: Arize AX / OTel setup + OpenAI auto-instrumentation
- src/register-prompt.ts: optional — publish the load_page tool to the AX
  Prompt Hub via @arizeai/ax-client

Verified end to end against live Arize AX, Browserbase, and OpenAI.

Author: Nancy-Chauhan

README.md

@@ -24,6 +24,7 @@ integrations/
 │   └── integrations/
 │       ├── agentkit/             # AgentKit implementations
 │       ├── agno/                 # Agno agent framework integration
+│       ├── arize/                # Arize AX observability & tracing
 │       ├── braintrust/           # Evaluation and testing tools
 │       ├── browser-use/          # Browser Use automation
 │       ├── cartesia/             # Cartesia integration

examples/integrations/arize/.env.example

@@ -0,0 +1,11 @@
+# Arize AX credentials (Space Settings → copy Space ID and API Key)
+ARIZE_SPACE_ID=
+ARIZE_API_KEY=
+ARIZE_PROJECT_NAME=browserbase-arize-agent
+
+# Browserbase credentials
+BROWSERBASE_API_KEY=
+
+# OpenAI (drives the agent)
+OPENAI_API_KEY=
+OPENAI_MODEL=gpt-5.4   # optional, defaults to gpt-5.4

examples/integrations/arize/README.md

@@ -0,0 +1,59 @@
+# Arize AX Integration
+
+Use Browserbase as a **tool** for an agent, and trace every tool call with [Arize AX](https://arize.com/docs/ax).
+
+## Overview
+
+This integration makes Browserbase a **tool** an agent can call — here a `load_page` tool that fetches a page's content as markdown via the Browserbase [Fetch API](https://www.browserbase.com/search) (no full browser needed). The agent uses tool calling to perform a task (e.g. read a page and summarize it), and **Arize AX traces the whole run** so you can debug, monitor, and evaluate it.
+
+Arize AX is an observability and evaluation platform for AI applications. It ingests [OpenTelemetry](https://opentelemetry.io/) traces via [OpenInference](https://github.com/Arize-ai/openinference), so each part of the agent shows up as a span:
+
+- **LLM spans** — every OpenAI chat completion (captured automatically), with token usage and cost
+- **Tool spans** — the Browserbase `load_page` (Fetch API) call, with the URL in and page markdown out
+- **Chain span** — groups the whole agent run; token usage and cost roll up to the trace
+
+## Setup
+
+Requires **Node.js 18.19+** (or 20.6+) — the OpenTelemetry dependencies set this baseline.
+
+```bash
+npm install
+cp .env.example .env   # then fill in your keys
+```
+
+| Variable                 | Where to find it                                 |
+| ------------------------ | ------------------------------------------------ |
+| `ARIZE_SPACE_ID`         | Arize AX → Space Settings                        |
+| `ARIZE_API_KEY`          | Arize AX → Space Settings                        |
+| `ARIZE_PROJECT_NAME`     | Any name; groups traces in AX (optional)         |
+| `BROWSERBASE_API_KEY`    | [Browserbase dashboard](https://browserbase.com) |
+| `OPENAI_API_KEY`         | OpenAI dashboard                                 |
+
+## Run
+
+```bash
+npm start
+# or ask your own question:
+npm start "Read https://docs.browserbase.com and list the main products."
+```
+
+The agent browses with Browserbase, prints an answer, and the trace appears in your Arize AX project as a `CHAIN → LLM → TOOL(load_page)` tree.
+
+## Files
+
+- [`src/instrumentation.ts`](src/instrumentation.ts) — Arize AX / OpenTelemetry setup (OTLP exporter, OpenAI instrumentation). Must be imported before `openai`.
+- [`src/agent.ts`](src/agent.ts) — the tool-calling agent: a loop over the Browserbase `load_page` tool, with manual `CHAIN` and `TOOL` spans.
+
+## Optional: register the tool in the Prompt Hub
+
+To also publish the `load_page` tool as a prompt in the Arize AX Prompt Hub (so you can iterate on prompting + tool calling there):
+
+```bash
+npm run register-prompt
+```
+
+This uses [`@arizeai/ax-client`](https://www.npmjs.com/package/@arizeai/ax-client) to create a `browserbase-load-page` prompt carrying the tool schema. AX stores the schema; `agent.ts` is what executes the tool.
+
+---
+
+You've integrated Browserbase with Arize AX! Use Browserbase as a tool to build composable agents that perform tasks like data extraction and RAG — and let Arize AX trace and evaluate every prompt, tool call, and token.