vercel-ai-sdk 0.0.1.dev1__tar.gz

vercel_ai_sdk-0.0.1.dev1/PKG-INFO

@@ -0,0 +1,128 @@
+Metadata-Version: 2.3
+Name: vercel-ai-sdk
+Version: 0.0.1.dev1
+Summary: The AI Toolkit for Python
+Author: Andrey Buzin
+Author-email: Andrey Buzin <andrey.buzin@vercel.com>
+Requires-Dist: anthropic>=0.40.0
+Requires-Dist: httpx>=0.28.1
+Requires-Dist: mcp>=1.18.0
+Requires-Dist: openai>=2.14.0
+Requires-Dist: pydantic>=2.12.5
+Requires-Python: >=3.12
+Description-Content-Type: text/markdown
+
+# ai-sdk
+
+A Python version of the [AI SDK](https://ai-sdk.dev/).
+
+## Quick Start
+
+```bash
+uv add ai-sdk
+```
+
+```python
+import ai_sdk as ai
+
+llm = ai.openai.OpenAIModel(model="gpt-4", api_key="...")
+
+async for msg in ai.execute(my_agent, llm, "Hello"):
+    print(msg.text)
+```
+
+## API Reference
+
+### `@ai.tool`
+
+Decorator that turns an async function into a tool. Parameters are auto-extracted from type hints and docstrings become the tool description.
+
+```python
+@ai.tool
+async def get_weather(city: str, units: str = "celsius") -> str:
+    """Get current weather for a city."""
+    return f"72°F in {city}"
+```
+
+### `ai.stream_text(llm, messages, label=None)`
+
+Streams text from the LLM without tool support. Returns a `Stream` that can be awaited or async-iterated.
+
+```python
+result = await ai.stream_text(llm, messages)
+# or iterate for real-time updates
+async for msg in ai.stream_text(llm, messages):
+    print(msg.text_delta, end="")
+```
+
+### `ai.stream_loop(llm, messages, tools, label=None)`
+
+Streams LLM responses and automatically executes tool calls in a loop until complete. This is the main function for agentic workflows.
+
+```python
+result = await ai.stream_loop(llm, messages, tools=[get_weather])
+```
+
+### `ai.execute(root_fn, *args)`
+
+Runs an agent function and yields all messages from nested streams. Use this as the top-level entry point for any agent workflow.
+
+```python
+async def my_agent(llm, query):
+    return await ai.stream_loop(llm, messages, tools)
+
+async for msg in ai.execute(my_agent, llm, "What's the weather?"):
+    print(msg)
+```
+
+### `ai.Message`
+
+Universal message type with `role` ("user", "assistant", "system") and `parts`. Access text via `msg.text`. The `label` field tags messages for multi-agent routing.
+
+### `ai.TextPart`, `ai.ToolPart`, `ai.ReasoningPart`
+
+Message parts. `TextPart` holds text content. `ToolPart` contains tool invocation details and results. `ReasoningPart` holds model reasoning/thinking output.
+
+## MCP Integration
+
+### `ai.mcp.get_http_tools(url, headers={}, tool_prefix="")`
+
+Connects to an MCP server over HTTP and returns tools. Optional `tool_prefix` namespaces tool names.
+
+```python
+tools = await ai.mcp.get_http_tools(
+    "https://mcp.example.com/mcp",
+    headers={"API_KEY": "..."},
+    tool_prefix="docs"
+)
+```
+
+### `ai.mcp.get_stdio_tools(cmd, *args, tool_prefix="")`
+
+Spawns an MCP server process via stdio. Useful for local MCP servers like npx packages.
+
+```python
+tools = await ai.mcp.get_stdio_tools(
+    "npx", "-y", "@upstash/context7-mcp",
+    "--api-key", os.environ["CONTEXT7_API_KEY"],
+    tool_prefix="context7"
+)
+```
+
+## Multi-Agent Example
+
+```python
+async def multiagent(llm, query):
+    # Run two agents in parallel
+    stream1, stream2 = await asyncio.gather(
+        ai.stream_loop(llm, msgs1, tools=[add_one], label="agent1"),
+        ai.stream_loop(llm, msgs2, tools=[multiply], label="agent2"),
+    )
+
+    # Combine results and summarize
+    combined = stream1[-1].text + stream2[-1].text
+    return await ai.stream_text(llm, make_messages(combined), label="summarizer")
+
+async for msg in ai.execute(multiagent, llm, "10"):
+    print(f"[{msg.label}] {msg.text_delta}", end="")
+```

vercel_ai_sdk-0.0.1.dev1/README.md

@@ -0,0 +1,114 @@
+# ai-sdk
+
+A Python version of the [AI SDK](https://ai-sdk.dev/).
+
+## Quick Start
+
+```bash
+uv add ai-sdk
+```
+
+```python
+import ai_sdk as ai
+
+llm = ai.openai.OpenAIModel(model="gpt-4", api_key="...")
+
+async for msg in ai.execute(my_agent, llm, "Hello"):
+    print(msg.text)
+```
+
+## API Reference
+
+### `@ai.tool`
+
+Decorator that turns an async function into a tool. Parameters are auto-extracted from type hints and docstrings become the tool description.
+
+```python
+@ai.tool
+async def get_weather(city: str, units: str = "celsius") -> str:
+    """Get current weather for a city."""
+    return f"72°F in {city}"
+```
+
+### `ai.stream_text(llm, messages, label=None)`
+
+Streams text from the LLM without tool support. Returns a `Stream` that can be awaited or async-iterated.
+
+```python
+result = await ai.stream_text(llm, messages)
+# or iterate for real-time updates
+async for msg in ai.stream_text(llm, messages):
+    print(msg.text_delta, end="")
+```
+
+### `ai.stream_loop(llm, messages, tools, label=None)`
+
+Streams LLM responses and automatically executes tool calls in a loop until complete. This is the main function for agentic workflows.
+
+```python
+result = await ai.stream_loop(llm, messages, tools=[get_weather])
+```
+
+### `ai.execute(root_fn, *args)`
+
+Runs an agent function and yields all messages from nested streams. Use this as the top-level entry point for any agent workflow.
+
+```python
+async def my_agent(llm, query):
+    return await ai.stream_loop(llm, messages, tools)
+
+async for msg in ai.execute(my_agent, llm, "What's the weather?"):
+    print(msg)
+```
+
+### `ai.Message`
+
+Universal message type with `role` ("user", "assistant", "system") and `parts`. Access text via `msg.text`. The `label` field tags messages for multi-agent routing.
+
+### `ai.TextPart`, `ai.ToolPart`, `ai.ReasoningPart`
+
+Message parts. `TextPart` holds text content. `ToolPart` contains tool invocation details and results. `ReasoningPart` holds model reasoning/thinking output.
+
+## MCP Integration
+
+### `ai.mcp.get_http_tools(url, headers={}, tool_prefix="")`
+
+Connects to an MCP server over HTTP and returns tools. Optional `tool_prefix` namespaces tool names.
+
+```python
+tools = await ai.mcp.get_http_tools(
+    "https://mcp.example.com/mcp",
+    headers={"API_KEY": "..."},
+    tool_prefix="docs"
+)
+```
+
+### `ai.mcp.get_stdio_tools(cmd, *args, tool_prefix="")`
+
+Spawns an MCP server process via stdio. Useful for local MCP servers like npx packages.
+
+```python
+tools = await ai.mcp.get_stdio_tools(
+    "npx", "-y", "@upstash/context7-mcp",
+    "--api-key", os.environ["CONTEXT7_API_KEY"],
+    tool_prefix="context7"
+)
+```
+
+## Multi-Agent Example
+
+```python
+async def multiagent(llm, query):
+    # Run two agents in parallel
+    stream1, stream2 = await asyncio.gather(
+        ai.stream_loop(llm, msgs1, tools=[add_one], label="agent1"),
+        ai.stream_loop(llm, msgs2, tools=[multiply], label="agent2"),
+    )
+
+    # Combine results and summarize
+    combined = stream1[-1].text + stream2[-1].text
+    return await ai.stream_text(llm, make_messages(combined), label="summarizer")
+
+async for msg in ai.execute(multiagent, llm, "10"):
+    print(f"[{msg.label}] {msg.text_delta}", end="")
+```

vercel_ai_sdk-0.0.1.dev1/pyproject.toml

@@ -0,0 +1,26 @@
+[project]
+name = "vercel-ai-sdk"
+version = "0.0.1.dev1"
+description = "The AI Toolkit for Python"
+readme = "README.md"
+authors = [
+    { name = "Andrey Buzin", email = "andrey.buzin@vercel.com" }
+]
+requires-python = ">=3.12"
+dependencies = [
+    "anthropic>=0.40.0",
+    "httpx>=0.28.1",
+    "mcp>=1.18.0",
+    "openai>=2.14.0",
+    "pydantic>=2.12.5",
+]
+
+[build-system]
+requires = ["uv_build>=0.9.21,<0.10.0"]
+build-backend = "uv_build"
+
+[dependency-groups]
+dev = [
+    "python-dotenv>=1.2.1",
+    "rich>=14.2.0",
+]

vercel_ai_sdk-0.0.1.dev1/src/vercel_ai_sdk/__init__.py

@@ -0,0 +1,42 @@
+from . import anthropic, mcp, openai, ai_sdk_ui
+
+# Re-export core types for convenient access
+from .core.messages import (
+    Message,
+    Part,
+    TextPart,
+    ToolPart,
+    ToolDelta,
+    ReasoningPart,
+)
+from .core.tools import Tool, tool
+from .core.runtime import (
+    LanguageModel,
+    Stream,
+    execute,
+    stream_loop,
+    stream_text,
+)
+
+__all__ = [
+    # Types
+    "Message",
+    "Part",
+    "TextPart",
+    "ToolPart",
+    "ToolDelta",
+    "ReasoningPart",
+    "Tool",
+    "Stream",
+    "LanguageModel",
+    # Functions
+    "tool",
+    "execute",
+    "stream_loop",
+    "stream_text",
+    # Submodules
+    "anthropic",
+    "mcp",
+    "openai",
+    "ai_sdk_ui",
+]

vercel_ai_sdk-0.0.1.dev1/src/vercel_ai_sdk/ai_sdk_ui/__init__.py

@@ -0,0 +1,26 @@
+from .adapter import (
+    # Internal → UI stream conversion
+    to_ui_message_stream,
+    to_sse_stream,
+    # UI → Internal message conversion
+    to_messages,
+    UIMessage,
+    UIMessagePart,
+    UITextPart,
+    UIReasoningPart,
+    UIToolInvocationPart,
+    # Headers for streaming responses
+    UI_MESSAGE_STREAM_HEADERS,
+)
+
+__all__ = [
+    "to_ui_message_stream",
+    "to_sse_stream",
+    "to_messages",
+    "UIMessage",
+    "UIMessagePart",
+    "UITextPart",
+    "UIReasoningPart",
+    "UIToolInvocationPart",
+    "UI_MESSAGE_STREAM_HEADERS",
+]