@starroy/ai-ping-core

Reusable protocol checks for AI and LLM API endpoints.

Supported profiles

| Profile | API style | Authentication | Streaming | Alias | | --- | --- | --- | --- | --- | | openai-chat | OpenAI-compatible Chat Completions API | Usually requires an API key | SSE | openai | | openai-responses | OpenAI-compatible Responses API | Usually requires an API key | Semantic SSE events | | | ollama | Ollama local APIs | No API key by default | JSON lines | | | gemini | Gemini Developer API REST | x-goog-api-key | SSE | | | anthropic | Anthropic Claude Messages API | x-api-key + anthropic-version | SSE | |

The OpenAI-compatible Chat Completions profile currently includes:

• openai-chat.models.list
• openai-chat.chat.basic
• openai-chat.chat.stream
• openai-chat.tool_calls.basic
• openai-chat.tool_calls.stream
• openai-chat.error.format

The canonical profile id is openai-chat. The older openai profile id remains supported as a backward-compatible alias. This profile sends Chat Completions-style messages requests and validates choices[].message / choices[].delta responses.

The OpenAI-compatible tool call checks validate modern Chat Completions tools / tool_calls, including streaming delta.tool_calls argument assembly and JSON parsing. They are recommended checks; legacy function_call is detected but does not pass modern tool call compatibility.

The OpenAI-compatible Responses profile currently includes:

• openai-responses.models.list
• openai-responses.responses.basic
• openai-responses.responses.stream
• openai-responses.error.format

The canonical profile id is openai-responses. It covers POST /responses, sends Responses-style input requests, and validates output, output_text, and semantic streaming events such as response.output_text.delta. It does not check Responses tools, built-in tools, multimodal input, or conversation state in v0.10.

The Ollama profile currently includes:

• ollama.tags
• ollama.generate.basic
• ollama.generate.stream
• ollama.chat.basic
• ollama.chat.stream

This covers Ollama native /api/tags, /api/generate, and /api/chat. /api/generate is Ollama's prompt-style native API, while /api/chat is its messages-style native API. For Ollama's OpenAI-compatible /v1/chat/completions, use the openai-chat profile instead.

The Gemini profile currently includes:

• gemini.models.list
• gemini.generate.basic
• gemini.generate.stream
• gemini.error.format

This covers Gemini Developer API REST endpoints under a base URL such as https://generativelanguage.googleapis.com/v1beta. It uses x-goog-api-key authentication and SSE streaming with Gemini response chunks, not OpenAI delta chunks.

The Anthropic profile currently includes:

• anthropic.models.list
• anthropic.messages.basic
• anthropic.messages.stream
• anthropic.error.format

This covers Anthropic Claude Messages API endpoints under a base URL such as https://api.anthropic.com/v1. It uses x-api-key and anthropic-version: 2023-06-01. Anthropic streaming is event-based SSE, not OpenAI delta streaming. Tool use, extended thinking, Bedrock Anthropic, and Vertex AI Anthropic are not covered by this profile.

import { runChecks } from "@starroy/ai-ping-core";

const report = await runChecks({
  profile: "openai-chat",
  baseUrl: "http://localhost:3000/v1",
  apiKey: "sk-test",
  model: "gpt-5.5",
});

const report = await runChecks({
  profile: "openai-responses",
  baseUrl: "https://api.openai.com/v1",
  apiKey: process.env.OPENAI_API_KEY,
  model: "gpt-5.1-mini",
});

const report = await runChecks({
  profile: "ollama",
  baseUrl: "http://localhost:11434",
  model: "llama3.2",
});

const report = await runChecks({
  profile: "gemini",
  baseUrl: "https://generativelanguage.googleapis.com/v1beta",
  apiKey: process.env.GEMINI_API_KEY,
  model: "gemini-2.5-flash",
});

const report = await runChecks({
  profile: "anthropic",
  baseUrl: "https://api.anthropic.com/v1",
  apiKey: process.env.ANTHROPIC_API_KEY,
  model: "claude-sonnet-4-5",
});

Ollama does not require an API key. Its streaming responses use JSON lines, not SSE.

runChecks returns a structured report and does not print, write files, set exit codes, or manage UI state. Future CLI and desktop entry points should consume this package instead of reimplementing checks. The CLI can render the same structured report as console, JSON, or a static HTML report with aiping check --html reports/aiping.html.