Files
hermes-agent/agent/turn_request_assembly.py

288 lines
13 KiB
Python

"""Per-iteration API request assembly for the conversation turn loop: build ``api_messages``
from the transcript, append MoA context, inject prefills, run the context-engine selection
hook and the send-time sanitizers, canonicalize for bit-perfect cache prefixes, build the
request-local prompt-cache plan LAST (after every transcript mutation), prepare the
persistent-MoA request, then measure request pressure. Extracted from
``run_conversation``; nothing here imports ``agent.conversation_loop`` at module level
(cycle) — loop-internal helpers resolve lazily so ``patch("agent.conversation_loop.X")``
sites keep intercepting.
"""
from __future__ import annotations
from dataclasses import dataclass
import logging
from typing import Any, Dict, Optional
from agent.message_sanitization import _sanitize_messages_surrogates
from agent.model_metadata import anchored_context_tokens
from agent.prompt_caching import build_prompt_cache_plan, effective_cache_ttl
from agent.turn_context import build_api_messages
logger = logging.getLogger("agent.conversation_loop")
@dataclass
class AssembledRequest:
"""Always ``action == "fallthrough"``; the fields are the iteration locals the assembly
produces (``api_messages``/``tools_for_api`` are the decorated request copies — the
canonical ``messages``/``agent.tools`` stay undecorated)."""
action: str
api_messages: Any
tools_for_api: Any
_moa_prepared_request: Any
pending_moa_prepared_request: Any
approx_tokens: Any
request_pressure_tokens: Any
total_chars: Any
def assemble_api_request(
agent: Any,
*,
messages: Any,
current_turn_user_idx: Any,
_ext_prefetch_cache: Any,
_plugin_user_context: Any,
moa_config: Any,
active_system_prompt: Any,
original_user_message: Any,
pending_moa_prepared_request: Any,
request_logger: Any,
) -> AssembledRequest:
"""Assemble the request in the original order. ORDER IS LOAD-BEARING: cache breakpoints
are injected only after whitespace normalization, the orphan sweep, thinking-only drop /
user merge and surrogate stripping, so the same row's bytes never vary across turns."""
from agent.conversation_loop import (
_apply_context_engine_selection,
_canonicalize_api_tool_calls,
_clone_message_for_send,
_midturn_request_pressure_tokens,
estimate_messages_tokens_rough,
)
def _verdict(action: str, result: Optional[Dict[str, Any]] = None) -> AssembledRequest:
return AssembledRequest(
action=action,
api_messages=api_messages,
tools_for_api=tools_for_api,
_moa_prepared_request=_moa_prepared_request,
pending_moa_prepared_request=pending_moa_prepared_request,
approx_tokens=approx_tokens,
request_pressure_tokens=request_pressure_tokens,
total_chars=total_chars,
)
api_messages, effective_system = build_api_messages(
agent,
messages,
current_turn_user_idx=current_turn_user_idx,
ext_prefetch_cache=_ext_prefetch_cache,
plugin_user_context=_plugin_user_context,
moa_config=moa_config,
active_system_prompt=active_system_prompt,
)
if moa_config:
try:
from agent.message_content import flatten_message_text as _flatten_mt
from agent.moa_loop import _preset_temperature, aggregate_moa_context
_moa_context = aggregate_moa_context(
user_prompt=(
original_user_message
if isinstance(original_user_message, str)
# Multimodal content list: extract visible text rather than
# str()-ing parts, which would leak base64 image payloads.
else _flatten_mt(original_user_message)
),
api_messages=api_messages,
reference_models=moa_config.get("reference_models") or [],
aggregator=moa_config.get("aggregator") or {},
temperature=_preset_temperature(moa_config, "reference_temperature"),
aggregator_temperature=_preset_temperature(moa_config, "aggregator_temperature"),
reference_max_tokens=moa_config.get("reference_max_tokens"),
# None = no per-preset override; inherit
# auxiliary.moa_reference.timeout via call_llm.
reference_timeout=(
float(moa_config["reference_timeout"])
if moa_config.get("reference_timeout")
else None
),
degraded_reference_policy=str(
moa_config.get("degraded_reference_policy") or "loud"
),
agent=agent,
)
if _moa_context:
for _msg in reversed(api_messages):
if _msg.get("role") == "user":
_base = _msg.get("content", "")
if isinstance(_base, str):
_msg["content"] = _base + "\n\n" + _moa_context
elif isinstance(_base, list):
# Multimodal turn: append MoA context as a trailing text
# part instead of silently dropping it.
_msg["content"] = [
*_base,
{"type": "text", "text": "\n\n" + _moa_context},
]
break
except Exception as _moa_exc:
logger.warning("MoA context aggregation failed: %s", _moa_exc)
# Inject ephemeral prefill messages right after the system prompt
# but before conversation history. Same API-call-time-only pattern.
if agent.prefill_messages:
sys_offset = 1 if (api_messages and api_messages[0].get("role") == "system") else 0
for idx, pfm in enumerate(agent.prefill_messages):
# Structural clone: the in-place sanitizers below must not write
# through into agent.prefill_messages' nested containers.
api_messages.insert(sys_offset + idx, _clone_message_for_send(pfm))
# Per-turn context selection hook: an engine may select/replace context for THIS
# call only — request-only, fail-open, and independent of should_compress().
_sel_incoming = (
messages[current_turn_user_idx]
if 0 <= current_turn_user_idx < len(messages)
else None
)
api_messages = _apply_context_engine_selection(
agent,
api_messages,
messages,
_sel_incoming,
logger=request_logger,
)
# Runs unconditionally (not gated on context_compressor) so orphaned tool
# results from session loading or manual message edits are always caught.
api_messages = agent._sanitize_api_messages(api_messages)
# One-time repeated-heal notice goes out via the status/warning callback, NEVER
# appended to messages: the cached prompt prefix stays byte-identical (#96870).
try:
from agent.agent_runtime_helpers import (
consume_pending_sanitizer_heal_notice,
)
_heal_notice = consume_pending_sanitizer_heal_notice()
if _heal_notice:
agent._emit_warning(_heal_notice)
except Exception:
# A notice hiccup must never break the send path.
logger.debug("sanitizer heal notice delivery failed", exc_info=True)
# Drop thinking-only assistant turns + merge adjacent users, API copy only:
# Anthropic-style backends 400 on a trailing `thinking` block; history keeps it.
api_messages = agent._drop_thinking_only_and_merge_users(
api_messages,
drop_codex_reasoning_items=agent.api_mode != "codex_responses",
)
# Normalize whitespace and tool-call JSON for bit-perfect prefixes across turns
# (KV-cache reuse on local servers, better cloud cache hits); API copy only.
for am in api_messages:
if isinstance(am.get("content"), str):
am["content"] = am["content"].strip()
_canonicalize_api_tool_calls(api_messages)
# Strip lone surrogates (U+D800-U+DFFF) that some Ollama-served models emit;
# they crash json.dumps() inside the OpenAI SDK and trigger the 3-retry cycle.
_sanitize_messages_surrogates(api_messages)
# No send-time pad loop here: ``repair_empty_non_final_messages`` (inside
# ``_sanitize_api_messages``) is the single owner of empty-turn repair, and its
# non-whitespace placeholder survives normalization regardless of ordering.
# Build the request-local cache sections LAST, after every transcript mutation;
# the canonical tool registry stays undecorated. Marked ``content`` becomes text
# blocks the whitespace pass skips, so the same row's bytes vary across turns.
tools_for_api = agent.tools
if agent._use_prompt_caching and agent.provider != "moa":
from agent.prompt_caching import (
envelope_tool_part_cache_markers_supported,
)
_static_system_prefix = getattr(agent, "_cached_system_prompt_static", None)
_initial_cache_plan = build_prompt_cache_plan(
api_messages,
tools_for_api,
# Clamp per-destination: a configured 1h regresses to 5m on
# Qwen/Alibaba routes, whose context cache is 5m-only (#84733).
cache_ttl=effective_cache_ttl(
agent._cache_ttl,
provider=agent.provider,
model=agent.model,
),
native_anthropic=agent._use_native_cache_layout,
static_system_prefix=(
_static_system_prefix
if isinstance(_static_system_prefix, str)
else None
),
direct_native_tool_cache=agent._direct_native_anthropic_tool_cache_capability(),
# LiteLLM-style envelope routes forward part-level markers into
# tool_result.content[] → non-retryable 400 (#89886).
tool_part_markers=envelope_tool_part_cache_markers_supported(
getattr(agent, "provider", ""), getattr(agent, "base_url", "")
),
)
api_messages = _initial_cache_plan.messages
tools_for_api = _initial_cache_plan.tools
# Prepare the persistent-MoA request before measuring compression pressure: the
# ephemeral advisor output is absent from ``messages``; ``create()`` reuses the
# prepared request instead of running the advisors again.
_moa_prepared_request = None
if agent.provider == "moa":
_moa_completions = getattr(getattr(agent.client, "chat", None), "completions", None)
if pending_moa_prepared_request is not None:
_rebase_moa_request = getattr(_moa_completions, "rebase_prepared_request", None)
if callable(_rebase_moa_request):
_moa_prepared_request = _rebase_moa_request(
pending_moa_prepared_request, api_messages
)
pending_moa_prepared_request = None
if _moa_prepared_request is None:
_prepare_moa_request = getattr(_moa_completions, "prepare", None)
if callable(_prepare_moa_request):
_moa_prepared_request = _prepare_moa_request(api_messages)
if _moa_prepared_request is not None:
api_messages = _moa_prepared_request["messages"]
# One image-stripped estimate feeds both figures; tools counted separately (50+
# tools ≈ 20-30K tokens); total_chars is a rough proxy for logs/hooks only.
# Charge stale thinking only when the active route replays it (#84371).
from agent.turn_context import _agent_stale_thinking_on_wire
if _agent_stale_thinking_on_wire(agent):
approx_tokens = estimate_messages_tokens_rough(api_messages)
else:
approx_tokens = estimate_messages_tokens_rough(
api_messages, charge_stale_thinking=False
)
# Route-aware: native Responses compaction prunes the wire payload, so the raw
# history figure overstates it and fires needless local compression (#96995).
request_pressure_tokens = _midturn_request_pressure_tokens(
agent, api_messages, effective_system or "", approx_tokens
)
# Usage-anchored override: real prompt_tokens (incl. system + tool schemas) +
# delta estimate replaces the whole-history heuristic when the anchor is fresh.
_anchored_pressure = anchored_context_tokens(
messages, getattr(agent, "_usage_anchor", None)
)
if _anchored_pressure is not None:
request_pressure_tokens = _anchored_pressure
total_chars = approx_tokens * 4
# Stash the rough estimate so update_from_response() can pair it with the real
# count (should_defer_preflight_to_real_usage). getattr: test doubles lack it.
_note_rough = getattr(
agent.context_compressor, "note_request_rough_estimate", None
)
if callable(_note_rough):
_note_rough(request_pressure_tokens)
return _verdict("fallthrough")