Delete ai.agent factory function

README.md

@@ -37,7 +37,7 @@ async def contact_mothership(query: str) -> str:
 
 async def main() -> None:
     model = ai.get_model("anthropic/claude-sonnet-4")
-    agent = ai.agent(tools=[contact_mothership])
+    agent = ai.Agent(tools=[contact_mothership])
 
     messages = [
         ai.system_message(

docs/ai-python/content/docs/basics/agents.mdx

@@ -30,7 +30,7 @@ async def contact_mothership(query: str) -> str:
 
 async def main() -> None:
     model = ai.get_model("anthropic/claude-sonnet-4")
-    agent = ai.agent(tools=[contact_mothership])
+    agent = ai.Agent(tools=[contact_mothership])
     messages = [
         ai.system_message(
             "Use the contact_mothership tool when asked about the future."

docs/ai-python/content/docs/basics/subagents-and-multi-agent.mdx

@@ -20,7 +20,7 @@ mothership_model = ai.get_model("anthropic/claude-sonnet-4")
 @ai.tool
 async def ask_mothership(topic: str) -> ai.SubAgentTool:
     """Ask a specialist agent for mothership guidance."""
-    sub_agent = ai.agent()
+    sub_agent = ai.Agent()
     sub_messages = [
         ai.system_message("Answer as the mothership operations desk."),
         ai.user_message(topic),

docs/ai-python/content/docs/basics/tools.mdx

@@ -167,7 +167,7 @@ tools = await ai.mcp.get_http_tools(
     headers={"Authorization": "Bearer your_access_token_here"},
 )
 
-agent = ai.agent(tools=tools)
+agent = ai.Agent(tools=tools)
 ```
 
 Use `ai.mcp.get_stdio_tools` for subprocess-based MCP servers.

docs/ai-python/content/docs/reference/ai/index.mdx

@@ -236,21 +236,19 @@ Part builder exports:
 
 ## Tools and Agents
 
-### agent
+### Agent
 
-`agent` creates an `Agent`.
+`Agent` runs the default agent loop.
 
 ```python
-agent = ai.agent(tools=[contact_mothership])
+agent = ai.Agent(tools=[contact_mothership])
 ```
 
 Arguments:
 
 - `tools`: Optional `AgentTool` values from `tool` and schema-only `Tool`
   declarations for provider-executed tools.
 
-Returns `Agent`.
-
 ### AgentTool
 
 `AgentTool` binds a model-facing `Tool` declaration to an executable Python
@@ -264,7 +262,7 @@ tool.validator
 tool.require_approval
 ```
 
-Pass `AgentTool` values to `agent(tools=[...])`.
+Pass `AgentTool` values to `Agent(tools=[...])`.
 
 ### Context
 

docs/ai-python/content/docs/reference/mcp.mdx

@@ -16,7 +16,7 @@ uv add "ai[mcp]"
 ## get_stdio_tools
 
 `get_stdio_tools` starts an MCP server subprocess and returns `AgentTool`
-values that can be passed to `ai.agent`.
+values that can be passed to `ai.Agent`.
 
 ```python
 tools = await ai.mcp.get_stdio_tools(
@@ -35,7 +35,7 @@ Arguments:
 ## get_http_tools
 
 `get_http_tools` connects to an MCP HTTP server and returns `AgentTool` values
-that can be passed to `ai.agent`.
+that can be passed to `ai.Agent`.
 
 ```python
 tools = await ai.mcp.get_http_tools("https://example.com/mcp")

skills/ai-python-basics/SKILL.md

@@ -59,7 +59,7 @@ async def get_weather(city: str) -> str:
 
 async def main() -> None:
     model = ai.get_model("anthropic/claude-sonnet-4")
-    agent = ai.agent(tools=[get_weather])
+    agent = ai.Agent(tools=[get_weather])
     messages = [
         ai.system_message("Use tools when useful."),
         ai.user_message("What is the weather in San Francisco?"),

skills/ai-python-streaming-tools/SKILL.md

@@ -50,7 +50,7 @@ Use `ai.SubAgentTool` when a tool runs another agent and streams its events.
 ```python
 @ai.tool
 async def research(topic: str) -> ai.SubAgentTool:
-    researcher = ai.agent()
+    researcher = ai.Agent()
 
     messages = [
         ai.system_message("Research briefly."),

skills/ai-python-subagents/SKILL.md

@@ -13,7 +13,7 @@ agent.
 ```python
 @ai.tool
 async def research(topic: str) -> ai.SubAgentTool:
-    child = ai.agent(tools=[lookup])
+    child = ai.Agent(tools=[lookup])
     messages = [
         ai.system_message("Research briefly."),
         ai.user_message(topic),

src/ai/__init__.py

@@ -10,7 +10,6 @@
     ToolCall,
     ToolRunner,
     abort_pending_hook,
-    agent,
     cancel_hook,
     hook,
     mcp,
@@ -176,7 +175,6 @@
     "UnsupportedProviderError",
     "VideoParams",
     "abort_pending_hook",
-    "agent",
     "assistant_message",
     "cancel_hook",
     "content_output",

src/ai/agents/__init__.py

@@ -18,7 +18,6 @@
     ToolCall,
     ToolCallCallable,
     ToolRunner,
-    agent,
     pending_tool_result,
     tool,
     tool_result,
@@ -53,7 +52,6 @@
     "ToolCallCallable",
     "ToolRunner",
     "abort_pending_hook",
-    "agent",
     "cancel_hook",
     "hook",
     "mcp",

src/ai/agents/agent.py

@@ -326,7 +326,7 @@ async def fetch(url: str) -> StreamingStatusTool[str]:
 
     @ai.tool
     async def research(topic: str) -> SubAgentTool:
-        sub = ai.agent(tools=[...])
+        sub = ai.Agent(tools=[...])
         async with sub.run(model, messages) as stream:
             async for event in stream:
                 yield event
@@ -1394,11 +1394,3 @@ async def _stream() -> AsyncGenerator[events_.AgentEvent]:
                     middleware_.deactivate(mw_token)
 
         yield AgentStream(_stream(), context)
-
-
-def agent(
-    *,
-    tools: Sequence[AgentTool | Tool] | None = None,
-) -> Agent:
-    """Create an Agent."""
-    return Agent(tools=tools)

src/ai/types/__init__.py

@@ -1,10 +1,9 @@
-from . import events, media, messages, proto, tools, usage
+from . import events, media, messages, tools, usage
 
 __all__ = [
     "events",
     "media",
     "messages",
-    "proto",
     "tools",
     "usage",
 ]

tests/agents/mcp/test_client.py

@@ -115,7 +115,7 @@ async def fake_fn(**kwargs: str) -> str:
     # Replace the executor (which would try to connect) with our fake.
     native = dataclasses.replace(native, fn=fake_fn)
 
-    my_agent = ai.agent(tools=[native])
+    my_agent = ai.Agent(tools=[native])
 
     call1 = [
         tool_call_msg(

tests/agents/test_aggregate_marker.py

@@ -188,7 +188,7 @@ async def alias_progress_tool(query: str) -> ai.StreamingStatusTool[str]:
 
 async def test_alias_declared_tool_runs_end_to_end() -> None:
     """An alias-declared streaming tool behaves like the kwarg form."""
-    my_agent = ai.agent(tools=[alias_progress_tool])
+    my_agent = ai.Agent(tools=[alias_progress_tool])
 
     call = [
         tool_call_msg(

tests/agents/test_generator_tools.py

@@ -38,7 +38,7 @@ async def progress_tool(query: str) -> AsyncGenerator[str]:
 async def test_generator_tool_streams_and_returns_result() -> None:
     """Generator tool yields streaming events visible to consumer;
     final text becomes the tool result fed back to the LLM."""
-    my_agent = ai.agent(tools=[progress_tool])
+    my_agent = ai.Agent(tools=[progress_tool])
 
     # Turn 1: LLM calls progress_tool
     # Turn 2: LLM produces final text after seeing the tool result
@@ -119,7 +119,7 @@ async def inner_fact(topic: str) -> str:
 @ai.tool(aggregator=ai.agents.MessageAggregator)
 async def research_tool(topic: str) -> AsyncGenerator[agent_events_.AgentEvent]:
     """Nested agent that researches a topic."""
-    inner = ai.agent(tools=[inner_fact])
+    inner = ai.Agent(tools=[inner_fact])
 
     msgs = [
         ai.system_message("Be concise."),
@@ -138,7 +138,7 @@ async def test_yield_from_nested_agent() -> None:
     queue and discards bare Messages, so the parent agent's
     context.messages stays clean.
     """
-    outer = ai.agent(tools=[research_tool])
+    outer = ai.Agent(tools=[research_tool])
 
     # Inner agent: text-only reply (no tools, to keep it simple)
     inner_reply = [text_msg("Mars has two moons.", id="inner-msg")]

tests/agents/test_run_blocked.py

@@ -57,7 +57,7 @@ def _multi_call_msg(*tc: tuple[str, str]) -> messages_.Message:
 
 async def test_gated_tool_block_cycle() -> None:
     """Park on approval -> RunBlocked; the resolved HookEvent unblocks."""
-    my_agent = ai.agent(tools=[gated])
+    my_agent = ai.Agent(tools=[gated])
     mock_llm(
         [
             [tool_call_msg(name="gated", args='{"x": 1}')],
@@ -118,7 +118,7 @@ async def slow(x: int) -> str:
         await release.wait()
         return "slow done"
 
-    my_agent = ai.agent(tools=[gated, slow])
+    my_agent = ai.Agent(tools=[gated, slow])
     mock_llm(
         [
             [_multi_call_msg(("tc-1", "gated"), ("tc-2", "slow"))],
@@ -154,7 +154,7 @@ async def slow(x: int) -> str:
 async def test_parallel_gated_tools() -> None:
     """Blocking needs *every* in-flight tool parked, and the signal
     re-fires when the run parks again on the remaining hook."""
-    my_agent = ai.agent(tools=[gated])
+    my_agent = ai.Agent(tools=[gated])
     mock_llm(
         [
             [_multi_call_msg(("tc-1", "gated"), ("tc-2", "gated"))],
@@ -203,7 +203,7 @@ async def loop(
 
 async def test_abort_leaves_blocked() -> None:
     """Serverless abort: the run ends still blocked, hooks still pending."""
-    my_agent = ai.agent(tools=[gated])
+    my_agent = ai.Agent(tools=[gated])
     mock_llm([[tool_call_msg(name="gated", args='{"x": 1}')]])
 
     blocked_events: list[events_.RunBlocked] = []
@@ -231,7 +231,7 @@ async def plain(x: int) -> str:
         """A plain tool."""
         return "plain done"
 
-    my_agent = ai.agent(tools=[plain])
+    my_agent = ai.Agent(tools=[plain])
     mock_llm(
         [
             [tool_call_msg(name="plain", args='{"x": 1}')],
@@ -260,7 +260,7 @@ async def test_block_signal_waits_for_stream_end() -> None:
             messages_.TextPart(text="still streaming..."),
         ],
     )
-    my_agent = ai.agent(tools=[gated])
+    my_agent = ai.Agent(tools=[gated])
     mock_llm([[msg], [text_msg("done", id="msg-2")]])
 
     delivered: list[events_.AgentEvent] = []
@@ -291,7 +291,7 @@ async def self_gated(x: int) -> str:
         await ai.hook("self_gate", payload=Confirmation)
         return "ran"
 
-    my_agent = ai.agent(tools=[self_gated])
+    my_agent = ai.Agent(tools=[self_gated])
     mock_llm(
         [
             [tool_call_msg(name="self_gated", args='{"x": 1}')],

tests/agents/test_runtime.py

@@ -33,7 +33,7 @@ async def concat(a: str, b: str) -> str:
 
 async def test_agent_text_only() -> None:
     """Agent default loop with no tool calls returns after one LLM call."""
-    my_agent = ai.agent(tools=[double])
+    my_agent = ai.Agent(tools=[double])
 
     llm = mock_llm([[text_msg("Hello!")]])
     async with my_agent.run(MOCK_MODEL, [ai.user_message("Hi")]) as stream:
@@ -47,7 +47,7 @@ async def test_agent_text_only() -> None:
 
 async def test_agent_tool_then_text() -> None:
     """Agent default loop calls tool, feeds result back, gets final text."""
-    my_agent = ai.agent(tools=[double])
+    my_agent = ai.Agent(tools=[double])
 
     call1 = [tool_call_msg(tc_id="tc-1", name="double", args='{"x": 5}')]
     call2 = [text_msg("The answer is 10.")]
@@ -69,7 +69,7 @@ async def test_agent_tool_then_text() -> None:
 
 async def test_agent_parallel_tools() -> None:
     """LLM returns two tool calls in one message; both execute."""
-    my_agent = ai.agent(tools=[double])
+    my_agent = ai.Agent(tools=[double])
 
     two_tools = messages.Message(
         id="msg-1",
@@ -104,7 +104,7 @@ async def test_agent_parallel_tools() -> None:
 
 async def test_agent_multi_turn() -> None:
     """LLM calls a tool, then calls another tool, then returns text."""
-    my_agent = ai.agent(tools=[double, concat])
+    my_agent = ai.Agent(tools=[double, concat])
 
     turn1 = [
         tool_call_msg(

tests/agents/test_validation_error_approval.py

@@ -32,7 +32,7 @@ async def edit(path: str, edits: list[TextEdit]) -> str:
 async def test_invalid_args_with_approval_returns_error_result() -> None:
     """Invalid tool args on an approval-gated tool should produce an error
     tool result without firing the approval hook."""
-    my_agent = ai.agent(tools=[edit])
+    my_agent = ai.Agent(tools=[edit])
 
     # Model sends edits as a dict instead of a list — this will fail validation
     bad_call = messages_.Message(
@@ -86,7 +86,7 @@ async def test_invalid_args_with_approval_returns_error_result() -> None:
 async def test_invalid_args_skips_approval_hook() -> None:
     """Invalid args should produce a validation error result without
     ever prompting for approval."""
-    my_agent = ai.agent(tools=[edit])
+    my_agent = ai.Agent(tools=[edit])
 
     bad_call = messages_.Message(
         id="msg-1",
@@ -132,7 +132,7 @@ async def test_invalid_args_skips_approval_hook() -> None:
 
 async def test_completely_invalid_json_with_approval() -> None:
     """Completely unparseable tool_args should also be handled gracefully."""
-    my_agent = ai.agent(tools=[edit])
+    my_agent = ai.Agent(tools=[edit])
 
     bad_call = messages_.Message(
         id="msg-1",