Skip to content

Your lightweight open-source personal agent.

refactor(memory): simplify idle consolidation flow

chengyongrufa0605abd0c5
diff --git a/nanobot/agent/memory.py b/nanobot/agent/memory.py
index 5ac841d81..92db1b2f1 100644
--- a/nanobot/agent/memory.py
+++ b/nanobot/agent/memory.py
@@ -1000,74 +1000,58 @@ class Consolidator:
runtime: LLMRuntime,
session_key: str | None = None,
summary_messages: list[dict[str, Any]] | None = None,
+ request_messages: list[dict[str, Any]] | None = None,
+ request_tools: list[dict[str, Any]] | None = None,
) -> str | None:
"""Summarize messages and append the result to history.jsonl.
``summary_messages`` adds context but is excluded from raw fallback.
+ ``request_messages`` preserves a prebuilt model-facing prefix instead
+ of flattening the messages; tools are included but disabled.
"""
if not messages:
return None
- messages_to_summarize = public_history_messages(
- summary_messages if summary_messages is not None else messages
- )
- formatted = MemoryStore._format_messages(messages_to_summarize)
- formatted = self._truncate_to_token_budget(formatted, runtime=runtime)
- system_prompt = render_template(
- "agent/consolidator_archive.md",
- strip=True,
- )
- return await self._archive_request(
- request_messages=[
+ prebuilt_request = request_messages is not None
+ if request_messages is None:
+ formatted = MemoryStore._format_messages(
+ public_history_messages(
+ summary_messages if summary_messages is not None else messages
+ )
+ )
+ formatted = self._truncate_to_token_budget(formatted, runtime=runtime)
+ request_messages = [
{
"role": "system",
- "content": system_prompt,
+ "content": render_template("agent/consolidator_archive.md", strip=True),
},
{"role": "user", "content": formatted},
- ],
- fallback_messages=messages,
- runtime=runtime,
- session_key=session_key,
- tools=None,
- tool_choice=None,
- )
-
- async def _archive_request(
- self,
- *,
- request_messages: list[dict[str, Any]],
- fallback_messages: list[dict[str, Any]],
- runtime: LLMRuntime,
- session_key: str | None,
- tools: list[dict[str, Any]] | None,
- tool_choice: str | dict[str, Any] | None,
- ) -> str | None:
- """Run one archive request and persist either its summary or a raw fallback."""
+ ]
try:
response = await runtime.provider.chat_with_retry(
model=runtime.model,
messages=request_messages,
- tools=tools,
- tool_choice=tool_choice,
+ tools=request_tools if prebuilt_request else None,
+ tool_choice="none" if prebuilt_request else None,
temperature=runtime.generation.temperature,
max_tokens=runtime.generation.max_tokens,
reasoning_effort=runtime.generation.reasoning_effort,
)
except Exception:
logger.warning("Consolidation provider call failed, raw-dumping to history")
- self.store.raw_archive(fallback_messages, session_key=session_key)
+ self.store.raw_archive(messages, session_key=session_key)
return None
if response.finish_reason == "error":
logger.warning("Consolidation provider returned an error, raw-dumping to history")
- self.store.raw_archive(fallback_messages, session_key=session_key)
+ self.store.raw_archive(messages, session_key=session_key)
return None
if response.has_tool_calls is True:
logger.warning("Consolidation provider returned tool calls, raw-dumping to history")
- self.store.raw_archive(fallback_messages, session_key=session_key)
+ self.store.raw_archive(messages, session_key=session_key)
return None
summary = response.content
if not summary or not summary.strip():
logger.warning("Consolidation provider returned no summary, raw-dumping to history")
- self.store.raw_archive(fallback_messages, session_key=session_key)
+ self.store.raw_archive(messages, session_key=session_key)
return None
self.store.append_history(
summary,
@@ -1090,33 +1074,42 @@ class Consolidator:
timestamp = last_active if isinstance(last_active, str) else session.updated_at.isoformat()
return f"Previous conversation summary (last active {timestamp}):\n{text}"
- def _build_idle_archive_messages(
+ async def _archive_idle_tail(
self,
session: Session,
messages: list[dict[str, Any]],
*,
runtime: LLMRuntime,
- ) -> list[dict[str, Any]] | None:
- """Append a temporary archive turn to the session's model-facing prefix."""
- replay_budget = self._input_token_budget(runtime)
- if replay_budget <= 0:
- replay_budget = max(128, runtime.context_window_tokens // 2)
+ ) -> str | None:
+ """Archive an idle tail by extending the ordinary model-facing messages."""
+ budget = self._input_token_budget(runtime)
+ if budget <= 0:
+ logger.debug(
+ "Idle consolidation has no safe input budget for {}; raw-dumping",
+ session.key,
+ )
+ self.store.raw_archive(messages, session_key=session.key)
+ return None
history = session.get_history(
max_messages=replay_max_messages_for_context(runtime.context_window_tokens),
- max_tokens=replay_budget,
+ max_tokens=budget,
)
archive_history = Session(
key=session.key,
messages=messages,
- ).get_history(max_messages=max(1, len(messages)))
+ ).get_history(max_messages=len(messages))
if (
not archive_history
- or len(history) < len(archive_history)
or history[-len(archive_history):] != archive_history
):
+ logger.debug(
+ "Idle consolidation cannot replay the full tail for {}; raw-dumping",
+ session.key,
+ )
+ self.store.raw_archive(messages, session_key=session.key)
return None
prompt = render_template(
- "agent/consolidator_idle_archive.md",
+ "agent/consolidator_archive.md",
strip=True,
archive_count=len(archive_history),
)
@@ -1124,7 +1117,7 @@ class Consolidator:
workspace: Path | None = None
if self._resolve_prompt_context is not None:
channel, workspace = self._resolve_prompt_context(session)
- built = self._build_messages(
+ request_messages = self._build_messages(
history=history,
current_message=prompt,
channel=channel,
@@ -1133,45 +1126,6 @@ class Consolidator:
session_key=session.key,
unified_session=self.unified_session,
)
- system_prefix = (
- [built[0]]
- if built and built[0].get("role") == "system"
- else []
- )
- return [
- *system_prefix,
- *history,
- {"role": "user", "content": prompt},
- ]
-
- async def _archive_idle_tail(
- self,
- session: Session,
- messages: list[dict[str, Any]],
- *,
- runtime: LLMRuntime,
- ) -> str | None:
- """Archive an idle tail by extending the ordinary model-facing messages."""
- request_messages = self._build_idle_archive_messages(
- session,
- messages,
- runtime=runtime,
- )
- if request_messages is None:
- logger.debug(
- "Idle consolidation cannot replay the full tail for {}; raw-dumping",
- session.key,
- )
- self.store.raw_archive(messages, session_key=session.key)
- return None
- budget = self._input_token_budget(runtime)
- if budget <= 0:
- logger.debug(
- "Idle consolidation has no safe input budget for {}; raw-dumping",
- session.key,
- )
- self.store.raw_archive(messages, session_key=session.key)
- return None
tools = self._get_tool_definitions()
estimated, source = estimate_prompt_tokens_chain(
runtime.provider,
@@ -1189,13 +1143,12 @@ class Consolidator:
)
self.store.raw_archive(messages, session_key=session.key)
return None
- return await self._archive_request(
- request_messages=request_messages,
- fallback_messages=messages,
+ return await self.archive(
+ messages,
runtime=runtime,
session_key=session.key,
- tools=tools,
- tool_choice="none",
+ request_messages=request_messages,
+ request_tools=tools,
)
async def maybe_consolidate_by_tokens(
diff --git a/nanobot/templates/agent/consolidator_archive.md b/nanobot/templates/agent/consolidator_archive.md
index 688e3012f..0f9a1d7c1 100644
--- a/nanobot/templates/agent/consolidator_archive.md
+++ b/nanobot/templates/agent/consolidator_archive.md
@@ -1,4 +1,10 @@
+{% if archive_count is defined %}
+Create a memory overview for only the final {{ archive_count }} conversation messages immediately before this instruction. Earlier messages are context for resolving references; do not summarize them again.
+
+Do not call tools. Return only the overview, following these memory rules:
+{% else %}
Extract key facts from this conversation. For each fact, annotate its memory attributes.
+{% endif %}
Only SNIP facts deserve a non-[skip] mark:
- Signal: would the user need to repeat this if forgotten?
diff --git a/nanobot/templates/agent/consolidator_idle_archive.md b/nanobot/templates/agent/consolidator_idle_archive.md
deleted file mode 100644
index 7ca46d345..000000000
--- a/nanobot/templates/agent/consolidator_idle_archive.md
+++ /dev/null
@@ -1,5 +0,0 @@
-Create a memory overview for only the final {{ archive_count }} conversation messages immediately before this instruction. Earlier messages are context for resolving references; do not summarize them again.
-
-Do not call tools. Return only the overview, following these memory rules:
-
-{% include 'agent/consolidator_archive.md' %}
diff --git a/tests/agent/test_consolidator.py b/tests/agent/test_consolidator.py
index 1a363da15..cef7420c5 100644
--- a/tests/agent/test_consolidator.py
+++ b/tests/agent/test_consolidator.py
@@ -274,6 +274,18 @@ class TestConsolidatorPromptContract:
assert "check context below" not in prompt.lower()
assert "Do not mark something [skip] merely because it might already exist" in prompt
+ def test_archive_prompt_scopes_idle_overview_when_message_count_is_provided(self):
+ prompt = render_template(
+ "agent/consolidator_archive.md",
+ strip=True,
+ archive_count=4,
+ )
+
+ assert "only the final 4 conversation messages" in prompt
+ assert "Earlier messages are context" in prompt
+ assert "Do not call tools" in prompt
+ assert "Only SNIP facts" in prompt
+
class TestConsolidatorArchiveErrorHandling:
"""archive() must fall back to raw_archive when the LLM returns an error