core/agent+cli: separate last-request usage from billed turn totals

Tool-loop rounds re-send history, so summed prompt tokens are billing
not context size. Track last_request_usage and label turn/session as
billed; clarify /status and end-of-turn lines.

CLAUDE.md

@@ -58,7 +58,7 @@ Async SQLAlchemy + aiosqlite. `MemoryStore`: schema init (+ lightweight SQLite `
 - **`ChatAgent`**: optional `MemoryStore` (persist on success only); `stream`; system prompt; `pending_retry_text` + `retry()`.
 - **`/compact`**: soft-compact tool dumps → model summary (no tools) → **new** session seeded with summary message.
 - Events: text_delta, assistant_message, tool_call/result, max_rounds, **error** (`retryable`/`source`), **cancelled** (`reason`), **usage** (`TokenUsage`).
-- Usage: API `usage` from completions (stream with `include_usage`); **char≈token fallback** (~4 chars/token) when omitted; `ChatAgent.session_usage` / `last_turn_usage`; CLI end-of-turn + `/status` (marks `(est)` / `(api+est)`).
+- Usage: API `usage` from completions (stream with `include_usage`); **char≈token fallback** (~4 chars/token) when omitted; `last_request_usage` (one model call), `last_turn_usage` / `session_usage` (**billed sums** — tool rounds re-send history); CLI end-of-turn + `/status`.
 - Config ``[agent]``: `system_prompt`, `max_tool_result_chars`, `parallel_tools`, `confirm_destructive`, `path_denylist`, `max_context_tokens` (default 200k est. tokens).
 
 ### Tools (`tools/`)

src/plyngent/agent/chat.py

@@ -42,6 +42,8 @@ class ChatAgent:
     pending_retry_text: str | None
     session_usage: TokenUsage
     last_turn_usage: TokenUsage
+    last_request_usage: TokenUsage
+    last_turn_rounds: int
 
     def __init__(
         self,
@@ -78,6 +80,8 @@ class ChatAgent:
         self.pending_retry_text = None
         self.session_usage = TokenUsage()
         self.last_turn_usage = TokenUsage()
+        self.last_request_usage = TokenUsage()
+        self.last_turn_rounds = 0
         self._ensure_system_prompt()
 
     def _ensure_system_prompt(self) -> None:
@@ -123,6 +127,8 @@ class ChatAgent:
 
         completed = False
         turn_usage = TokenUsage()
+        turn_rounds = 0
+        last_request = TokenUsage()
         try:
             async for event in run_chat_loop(
                 self.client,
@@ -138,6 +144,9 @@ class ChatAgent:
                 max_context_tokens=self.max_context_tokens,
             ):
                 if isinstance(event, UsageEvent):
+                    # Each tool-loop round re-sends history; sum is billing, not context size.
+                    turn_rounds += 1
+                    last_request = event.usage
                     turn_usage = turn_usage.add(event.usage)
                     self.session_usage = self.session_usage.add(event.usage)
                 yield event
@@ -148,6 +157,8 @@ class ChatAgent:
             raise
 
         self.last_turn_usage = turn_usage
+        self.last_request_usage = last_request
+        self.last_turn_rounds = turn_rounds
         for message in self.messages[pre_len:]:
             await self._persist(message)
         self.pending_retry_text = None

src/plyngent/agent/usage.py

@@ -40,15 +40,21 @@ class TokenUsage(Struct, omit_defaults=True):
     def is_zero(self) -> bool:
         return self.prompt_tokens == 0 and self.completion_tokens == 0 and self.total_tokens == 0
 
-    def format_line(self) -> str:
+    def format_line(self, *, billed: bool = False) -> str:
+        """Format for display.
+
+        When ``billed`` is True, label as cumulative API billing (sum of rounds),
+        not a single context snapshot — each tool-loop round re-sends history.
+        """
         tag = ""
         if self.source == "estimate":
             tag = " (est)"
         elif self.source == "mixed":
             tag = " (api+est)"
+        prefix = "billed " if billed else ""
         return (
-            f"tokens prompt={self.prompt_tokens} completion={self.completion_tokens} "
-            f"total={self.total_tokens}{tag}"
+            f"{prefix}tokens prompt={self.prompt_tokens} "
+            f"completion={self.completion_tokens} total={self.total_tokens}{tag}"
         )
 
 

src/plyngent/cli/repl.py

@@ -63,17 +63,22 @@ def _cmd_status(state: ReplState) -> None:
     ctx_budget = state.agent.max_context_tokens
     session_u = state.agent.session_usage
     last_u = state.agent.last_turn_usage
+    last_req = state.agent.last_request_usage
+    last_rounds = state.agent.last_turn_rounds
     click.echo(
         f"provider={state.provider_name}  model={state.model}\n"
         f"session={state.session_id}  messages={len(state.agent.messages)}  "
         f"pending_retry={pending_disp}\n"
         f"tools={'on' if state.tools_enabled else 'off'}  "
         f"rounds={state.max_rounds}  stream={'on' if state.agent.stream else 'off'}\n"
-        f"context_tokens~={ctx_tokens}/{ctx_budget} (est)  "
+        f"context_tokens~={ctx_tokens}/{ctx_budget} (est, once)  "
         f"context_chars={ctx_chars}  "
         f"tool_result_max={state.agent.max_tool_result_chars}\n"
-        f"usage_session={session_u.format_line()}\n"
-        f"usage_last_turn={last_u.format_line()}\n"
+        f"last_request={last_req.format_line()}  "
+        f"(last model call; ~context size if from API)\n"
+        f"usage_last_turn={last_u.format_line(billed=True)}  "
+        f"rounds={last_rounds}\n"
+        f"usage_session={session_u.format_line(billed=True)}\n"
         f"workspace={state.workspace}"
     )
 

src/plyngent/cli/retry.py

@@ -89,8 +89,19 @@ async def run_cancellable[T](coro: Coroutine[object, object, T]) -> T:
 
 
 def _echo_turn_usage(agent: ChatAgent) -> None:
+    if agent.last_turn_usage.is_zero() and agent.last_request_usage.is_zero():
+        return
+    rounds = agent.last_turn_rounds
+    parts: list[str] = []
+    if not agent.last_request_usage.is_zero():
+        parts.append(f"last_request {agent.last_request_usage.format_line()}")
     if not agent.last_turn_usage.is_zero():
-        click.secho(f"[{agent.last_turn_usage.format_line()}]", fg="bright_black")
+        label = agent.last_turn_usage.format_line(billed=True)
+        if rounds > 1:
+            parts.append(f"turn {label} over {rounds} rounds")
+        else:
+            parts.append(f"turn {label}")
+    click.secho(f"[{'; '.join(parts)}]", fg="bright_black")
 
 
 async def _wait_for_retry(attempt: int, max_retries: int, wait: float) -> bool:

tests/test_agent/test_loop.py

@@ -230,12 +230,51 @@ async def test_chat_agent_accumulates_session_usage() -> None:
     agent = ChatAgent(client, model="m", stream=False)
     _ = [e async for e in agent.run("one")]
     assert agent.last_turn_usage.total_tokens == 6
+    assert agent.last_request_usage.total_tokens == 6
+    assert agent.last_turn_rounds == 1
     assert agent.session_usage.total_tokens == 6
     _ = [e async for e in agent.run("two")]
     assert agent.last_turn_usage.total_tokens == 10
+    assert agent.last_request_usage.total_tokens == 10
     assert agent.session_usage.total_tokens == 16
 
 
+async def test_chat_agent_turn_usage_sums_tool_rounds() -> None:
+    """Multi-round tool loop: turn usage is billing sum; last_request is final call."""
+
+    @tool
+    def ping() -> str:
+        return "pong"
+
+    registry = ToolRegistry([ping])
+    client = ScriptedClient(
+        [
+            _response(
+                AssistantChatMessage(
+                    content="",
+                    tool_calls=[
+                        AssistantFunctionToolCall(
+                            id="1",
+                            function=AssistantFunctionTool(name="ping", arguments="{}"),
+                        )
+                    ],
+                ),
+                usage={"prompt_tokens": 100, "completion_tokens": 5, "total_tokens": 105},
+            ),
+            _response(
+                AssistantChatMessage(content="done"),
+                usage={"prompt_tokens": 200, "completion_tokens": 3, "total_tokens": 203},
+            ),
+        ]
+    )
+    agent = ChatAgent(client, model="m", tools=registry, stream=False)
+    _ = [e async for e in agent.run("go")]
+    assert agent.last_turn_rounds == 2
+    assert agent.last_request_usage.prompt_tokens == 200
+    assert agent.last_turn_usage.prompt_tokens == 300
+    assert agent.last_turn_usage.total_tokens == 308
+
+
 async def test_stream_yields_deltas_incrementally() -> None:
     """Text deltas are yielded as chunks arrive, not only after the full stream."""
 

tests/test_agent/test_usage.py

@@ -97,3 +97,5 @@ def test_format_line_marks_estimate() -> None:
     assert "(est)" in line
     mixed = TokenUsage(prompt_tokens=1, completion_tokens=0, total_tokens=1, source="mixed").format_line()
     assert "(api+est)" in mixed
+    billed = TokenUsage(prompt_tokens=1, completion_tokens=0, total_tokens=1).format_line(billed=True)
+    assert billed.startswith("billed tokens")