23301d705c
The 1h Anthropic cache tier writes at 2x base (5m: 1.25x) and only pays off when turns are more than five minutes apart. That is exactly the shape of an interactive session a person parks and resumes, and exactly not the shape of a subagent, cron run, one-shot or webhook that calls every few seconds and is gone. A single global `cache_ttl` cannot be right for both, so operators leave it on 5m and pay a full context re-write every time they come back to a CLI session after a coffee. Measured on one install (2 days of per-call API logs, Claude via the Nous route): 63% of interactive cache-write tokens were cold re-writes after a 5-60 minute idle gap; 1h would cut interactive write cost ~42% while costing ~49% more on subagents and ~23% more on cron. `auto` resolves once per session from the session source (`_session_source_for_agent`): 1h for cli/tui/desktop/messaging platforms, 5m for subagent, cron, oneshot, webhook, kanban, api. Auxiliary/stub calls keep 5m; the delegate_tool child clamp (#104168) still applies. Default stays "5m".
42 lines
2.2 KiB
Python
42 lines
2.2 KiB
Python
"""``prompt_caching.cache_ttl: auto`` picks the Anthropic cache tier by who paces the session.
|
|
|
|
The 1h tier writes at 2x base (5m: 1.25x) and only pays off across >5-minute pauses, i.e.
|
|
for a person; machine-paced sessions (subagent, cron, oneshot, webhook, ...) never collect the
|
|
retention. ``auto`` resolves once per session in ``_init_prompt_cache_config``.
|
|
"""
|
|
|
|
from agent.prompt_caching import MACHINE_PACED_SOURCES, auto_cache_ttl_for_source
|
|
|
|
|
|
def test_auto_tier_follows_session_pace():
|
|
assert auto_cache_ttl_for_source("cli") == "1h"
|
|
assert auto_cache_ttl_for_source("telegram") == "1h"
|
|
assert auto_cache_ttl_for_source(None) == "1h", "an unknown/blank source is a person until proven otherwise"
|
|
for source in MACHINE_PACED_SOURCES:
|
|
assert auto_cache_ttl_for_source(source) == "5m", source
|
|
|
|
|
|
def test_real_agent_resolves_auto_from_its_platform(tmp_path, monkeypatch):
|
|
"""A real AIAgent under ``cache_ttl: auto``: a CLI session lands on 1h, a cron session on 5m, and
|
|
caching itself stays enabled on both (auto is a tier choice, never a disable)."""
|
|
monkeypatch.setenv("HERMES_HOME", str(tmp_path))
|
|
monkeypatch.delenv("HERMES_SESSION_SOURCE", raising=False)
|
|
(tmp_path / "config.yaml").write_text(
|
|
"prompt_caching:\n cache_ttl: auto\nmodel:\n default: anthropic/claude-sonnet-4.6\n", encoding="utf-8")
|
|
from run_agent import AIAgent
|
|
kw = dict(api_key="k", base_url="https://openrouter.ai/api/v1", provider="openrouter",
|
|
api_mode="chat_completions", model="anthropic/claude-sonnet-4.6", quiet_mode=True,
|
|
skip_context_files=True, skip_memory=True, save_trajectories=False, enabled_toolsets=["file"])
|
|
interactive = AIAgent(session_id="p-cli", platform="cli", **kw)
|
|
scheduled = AIAgent(session_id="p-cron", platform="cron", **kw)
|
|
try:
|
|
assert interactive._cache_ttl == "1h"
|
|
assert scheduled._cache_ttl == "5m"
|
|
assert interactive._use_prompt_caching and scheduled._use_prompt_caching
|
|
assert not interactive._cache_disabled and not scheduled._cache_disabled
|
|
finally:
|
|
for a in (interactive, scheduled):
|
|
close = getattr(a, "close", None)
|
|
if callable(close):
|
|
close()
|