From 20573be38d97e80e0a41cf9b748795e35be80b42 Mon Sep 17 00:00:00 2001
From: "Michael J. Sullivan" <sully@msully.net>
Date: Tue, 27 Oct 2026 17:46:34 -0700
Subject: [PATCH] Add ai.durable.DurableAgent

Andrey and I are inclined towards *not* merging the loops.
---
 src/ai/__init__.py    |  2 +
 src/ai/durable.py     | 66 ++++++++++++++++++++++++++++++
 tests/test_durable.py | 93 +++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 161 insertions(+)
 create mode 100644 src/ai/durable.py
 create mode 100644 tests/test_durable.py

diff --git a/src/ai/__init__.py b/src/ai/__init__.py
index 171b6cdd..b10faf3e 100644
--- a/src/ai/__init__.py
+++ b/src/ai/__init__.py
@@ -1,4 +1,5 @@
 from . import (
+    durable,
     errors,
     experimental_telemetry,
     models,
@@ -195,6 +196,7 @@
     "current_tool_call",
     "defer_hook",
     "deferred_tool_result",
+    "durable",
     "errors",
     "events",
     "experimental_generate",
diff --git a/src/ai/durable.py b/src/ai/durable.py
new file mode 100644
index 00000000..f3735de2
--- /dev/null
+++ b/src/ai/durable.py
@@ -0,0 +1,66 @@
+from __future__ import annotations
+
+import contextlib
+from collections.abc import AsyncGenerator, Collection, Sequence
+from typing import Protocol
+
+from . import agents
+from .types import events
+
+__all__ = ["DurableAgent", "Streamer"]
+
+
+class Streamer(Protocol):
+    def __call__(
+        self, *, context: agents.Context
+    ) -> contextlib.AbstractAsyncContextManager[
+        AsyncGenerator[events.AgentEvent]
+    ]: ...
+
+
+class DurableAgent(agents.Agent):
+    # We require the loop to run in lockstep with the client code, so that
+    # tool streams are always sent before the next llm_step invocation.
+    LOOP_BUFFER = 0
+
+    def __init__(
+        self,
+        *,
+        tools: Sequence[agents.AgentTool | agents.Tool] | None = None,
+        streamer: Streamer,
+        eager_tools: Collection[str] = (),
+    ) -> None:
+        super().__init__(tools=tools)
+        self.streamer = streamer
+        self.eager_tools = frozenset(eager_tools)
+
+    async def loop(
+        self, context: agents.Context
+    ) -> AsyncGenerator[events.AgentEvent]:
+        while context.keep_running():
+            async with (
+                self.streamer(context=context) as stream,
+                agents.ToolRunner() as tr,
+            ):
+                final = False
+                async for event in stream:
+                    yield event
+
+                    if isinstance(event, events.ToolEnd):
+                        if (
+                            final
+                            or event.tool_call.tool_name in self.eager_tools
+                        ):
+                            tr.schedule(self.resolve(event.tool_call))
+                    elif isinstance(event, events.StreamEnd):
+                        tr.discard_all(
+                            except_=self.resolve(event.message.tool_calls)
+                        )
+                        context.add(event.message)
+                    elif isinstance(event, events.StreamStart):
+                        final = True
+
+                async for event in tr.events():
+                    yield event
+
+                context.add(tr.get_tool_message())
diff --git a/tests/test_durable.py b/tests/test_durable.py
new file mode 100644
index 00000000..4f0f41db
--- /dev/null
+++ b/tests/test_durable.py
@@ -0,0 +1,93 @@
+from __future__ import annotations
+
+import asyncio
+import contextlib
+from collections.abc import AsyncGenerator, AsyncIterator, Sequence
+
+import ai
+from ai import durable
+from ai.types import events, messages
+
+from .conftest import MOCK_MODEL
+
+log: list[str] = []
+
+
+@ai.tool
+async def double(x: int) -> int:
+    """Double a number."""
+    log.append(f"double:{x}")
+    return x * 2
+
+
+@ai.tool
+async def negate(x: int) -> int:
+    """Negate a number."""
+    log.append(f"negate:{x}")
+    return -x
+
+
+def scripted_streamer(
+    steps: list[tuple[Sequence[events.AgentEvent], messages.Message]],
+) -> durable.Streamer:
+    it = iter(steps)
+
+    @contextlib.asynccontextmanager
+    async def streamer(
+        *, context: ai.Context
+    ) -> AsyncIterator[AsyncGenerator[events.AgentEvent]]:
+        early, message = next(it)
+
+        async def stream() -> AsyncGenerator[events.AgentEvent]:
+            for event in early:
+                yield event
+            # stand-in for waiting on the llm step
+            for _ in range(10):
+                await asyncio.sleep(0)
+            log.append("replay")
+            async with ai.Stream.replay_message(message) as replay:
+                async for event in replay:
+                    yield event
+
+        yield stream()
+
+    return streamer
+
+
+async def test_durable_agent_eager_tools() -> None:
+    log.clear()
+    a = ai.testing.tool_call(double, x=1)
+    b = ai.testing.tool_call(negate, x=2)
+    dropped = ai.testing.tool_call(double, x=3)
+    streamer = scripted_streamer(
+        [
+            (
+                [
+                    events.ToolEnd(tool_call_id=a.tool_call_id, tool_call=a),
+                    events.ToolEnd(tool_call_id=b.tool_call_id, tool_call=b),
+                    events.ToolEnd(
+                        tool_call_id=dropped.tool_call_id, tool_call=dropped
+                    ),
+                ],
+                ai.assistant_message(a, b),
+            ),
+            ((), ai.assistant_message("done")),
+        ]
+    )
+    agent = durable.DurableAgent(
+        tools=[double, negate], streamer=streamer, eager_tools={"double"}
+    )
+
+    async with agent.run(MOCK_MODEL, [ai.user_message("go")]) as run:
+        async for _ in run:
+            pass
+
+    # eager tools start before the final message arrives and aren't rerun
+    assert log == ["double:1", "double:3", "replay", "negate:2", "replay"]
+    tool_message = run.messages[2]
+    assert {r.tool_call_id: r.result for r in tool_message.tool_results} == {
+        a.tool_call_id: 2,
+        b.tool_call_id: -2,
+    }
+    assert run.messages[-1].text == "done"
+    assert len(run.messages) == 4