mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-24 16:54:43 +00:00
72056faf8f
Long-lived sessions (e.g. a Telegram thread resumed over hours/days) accumulate a large context that the existing size-based threshold only trims once it crosses `threshold × context_window`. Until then every turn re-reads the full history, which on large-context models can mean hundreds of K of cache-read tokens per call even across long idle gaps. Add a time-based trigger that complements (does not replace) the size threshold: when a session resumes after `compression.idle_compact_after_seconds` of inactivity, compact the accumulated history up front, before the first reply. Disabled by default (0), so existing behaviour is unchanged. The trigger reuses `_last_activity_ts` (the last time the turn loop did work) to measure the idle gap at turn start, gates the token estimate behind a cheap gap pre-check, and skips compaction when the context is already at/below the post-compression target (threshold × target_ratio) so a short idle thread never pays for a summarization that saves nothing. It also defers to an active compression-failure cooldown. The decision is factored into a pure predicate, `_should_idle_compact`, which is unit-tested without a live agent.
55 lines
2 KiB
Python
55 lines
2 KiB
Python
"""Tests for the opt-in idle-triggered compaction policy.
|
|
|
|
Covers ``agent.turn_context._should_idle_compact`` — the pure predicate that
|
|
decides whether a session resuming after an idle gap should compact up front.
|
|
The predicate is intentionally side-effect-free so the policy can be verified
|
|
without constructing a live agent or DB.
|
|
"""
|
|
|
|
from agent.turn_context import _should_idle_compact
|
|
|
|
|
|
def _decide(**overrides):
|
|
"""Call the predicate with sensible defaults (idle + large context => fire)."""
|
|
kwargs = dict(
|
|
enabled=True,
|
|
idle_after_seconds=1800,
|
|
idle_gap_seconds=3600.0,
|
|
tokens=100_000,
|
|
floor_tokens=40_000,
|
|
cooldown_active=False,
|
|
)
|
|
kwargs.update(overrides)
|
|
return _should_idle_compact(**kwargs)
|
|
|
|
|
|
class TestShouldIdleCompact:
|
|
def test_fires_when_idle_long_enough_and_context_large(self):
|
|
assert _decide() is True
|
|
|
|
def test_disabled_when_idle_after_zero(self):
|
|
# 0 is the documented "off" value — must never fire regardless of gap.
|
|
assert _decide(idle_after_seconds=0, idle_gap_seconds=10_000.0) is False
|
|
|
|
def test_disabled_when_idle_after_negative(self):
|
|
assert _decide(idle_after_seconds=-1) is False
|
|
|
|
def test_disabled_when_compression_off(self):
|
|
assert _decide(enabled=False) is False
|
|
|
|
def test_skips_when_gap_below_threshold(self):
|
|
assert _decide(idle_gap_seconds=600.0) is False
|
|
|
|
def test_gap_exactly_at_threshold_fires(self):
|
|
assert _decide(idle_after_seconds=1800, idle_gap_seconds=1800.0) is True
|
|
|
|
def test_skips_when_context_at_or_below_floor(self):
|
|
# At/below the post-compression target there is nothing worth saving.
|
|
assert _decide(tokens=40_000, floor_tokens=40_000) is False
|
|
assert _decide(tokens=39_999, floor_tokens=40_000) is False
|
|
|
|
def test_fires_just_above_floor(self):
|
|
assert _decide(tokens=40_001, floor_tokens=40_000) is True
|
|
|
|
def test_defers_to_active_compression_cooldown(self):
|
|
assert _decide(cooldown_active=True) is False
|