seb/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/main' into bb/gui

Browse Source 
# Conflicts:
#	tui_gateway/server.py
This commit is contained in:
Brooklyn Nicholson
2026-05-13 07:37:05 -04:00
parent 49de1adc49 942adf6179
commit b6f2ff5136
37 changed files with 281 additions and 728 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/agent/test_portal_tags.py
+61
View File
@@ -0,0 +1,61 @@
"""Tests for agent.portal_tags — Nous Portal request tag contract."""
from __future__ import annotations
def test_hermes_client_tag_includes_current_version():
"""The client tag must reflect hermes_cli.__version__ verbatim."""
from hermes_cli import __version__
from agent.portal_tags import hermes_client_tag
assert hermes_client_tag() == f"client=hermes-client-v{__version__}"
def test_hermes_client_tag_format():
"""The client tag has the exact shape Nous Portal expects."""
from agent.portal_tags import hermes_client_tag
tag = hermes_client_tag()
assert tag.startswith("client=hermes-client-v")
# No spaces, no commas — single tag value
assert " " not in tag
assert "," not in tag
def test_nous_portal_tags_contains_product_and_client():
"""Every Nous Portal request gets BOTH the product tag and the version tag."""
from agent.portal_tags import hermes_client_tag, nous_portal_tags
tags = nous_portal_tags()
assert "product=hermes-agent" in tags
assert hermes_client_tag() in tags
assert len(tags) == 2
def test_nous_portal_tags_returns_fresh_list():
"""Callers mutate the returned list; we must not share state across calls."""
from agent.portal_tags import nous_portal_tags
a = nous_portal_tags()
a.append("client=test-mutation")
b = nous_portal_tags()
assert "client=test-mutation" not in b
def test_auxiliary_client_nous_extra_body_uses_helper():
"""auxiliary_client.NOUS_EXTRA_BODY must match the canonical helper output."""
from agent.auxiliary_client import NOUS_EXTRA_BODY
from agent.portal_tags import nous_portal_tags
assert NOUS_EXTRA_BODY == {"tags": nous_portal_tags()}
def test_nous_provider_profile_uses_helper():
"""The Nous provider profile (main agent loop) must use the canonical tags."""
from agent.portal_tags import nous_portal_tags
from providers import get_provider_profile
profile = get_provider_profile("nous")
assert profile is not None
body = profile.build_extra_body()
assert body["tags"] == nous_portal_tags()
tests/agent/test_prompt_caching.py
-131
View File
@@ -6,8 +6,6 @@ import pytest
from agent.prompt_caching import (
_apply_cache_marker,
apply_anthropic_cache_control,
apply_anthropic_cache_control_long_lived,
mark_tools_for_long_lived_cache,
)
@@ -143,132 +141,3 @@ class TestApplyAnthropicCacheControl:
elif "cache_control" in msg:
count += 1
assert count <= 4
class TestMarkToolsForLongLivedCache:
def test_returns_unchanged_for_empty_tools(self):
assert mark_tools_for_long_lived_cache(None) is None
assert mark_tools_for_long_lived_cache([]) == []
def test_marks_only_last_tool(self):
tools = [
{"type": "function", "function": {"name": "a"}},
{"type": "function", "function": {"name": "b"}},
{"type": "function", "function": {"name": "c"}},
]
out = mark_tools_for_long_lived_cache(tools)
assert "cache_control" not in out[0]
assert "cache_control" not in out[1]
assert out[2]["cache_control"] == {"type": "ephemeral", "ttl": "1h"}
def test_does_not_mutate_input(self):
tools = [{"type": "function", "function": {"name": "a"}}]
mark_tools_for_long_lived_cache(tools)
assert "cache_control" not in tools[0]
def test_5m_ttl_drops_ttl_field(self):
tools = [{"type": "function", "function": {"name": "a"}}]
out = mark_tools_for_long_lived_cache(tools, long_lived_ttl="5m")
assert out[0]["cache_control"] == {"type": "ephemeral"}
class TestApplyAnthropicCacheControlLongLived:
def test_empty_messages(self):
assert apply_anthropic_cache_control_long_lived([]) == []
def test_marks_first_block_of_split_system(self):
msgs = [
{"role": "system", "content": [
{"type": "text", "text": "STABLE"},
{"type": "text", "text": "CONTEXT"},
{"type": "text", "text": "VOLATILE"},
]},
{"role": "user", "content": "msg1"},
{"role": "assistant", "content": "msg2"},
]
out = apply_anthropic_cache_control_long_lived(msgs)
sys_blocks = out[0]["content"]
assert sys_blocks[0]["cache_control"] == {"type": "ephemeral", "ttl": "1h"}
assert "cache_control" not in sys_blocks[1]
assert "cache_control" not in sys_blocks[2]
def test_rolling_marker_on_last_2_messages(self):
msgs = [
{"role": "system", "content": [{"type": "text", "text": "S"}]},
{"role": "user", "content": "u1"},
{"role": "assistant", "content": "a1"},
{"role": "user", "content": "u2"},
{"role": "assistant", "content": "a2"},
]
out = apply_anthropic_cache_control_long_lived(msgs)
def has_marker(m):
c = m.get("content")
if isinstance(c, list) and c and isinstance(c[-1], dict):
return "cache_control" in c[-1]
return "cache_control" in m
# u1 and a1 (older messages) should NOT be marked
assert not has_marker(out[1])
assert not has_marker(out[2])
# u2 and a2 (last 2) SHOULD be marked
assert has_marker(out[3])
assert has_marker(out[4])
def test_rolling_marker_uses_5m_ttl(self):
msgs = [
{"role": "system", "content": [{"type": "text", "text": "S"}]},
{"role": "user", "content": "u1"},
{"role": "assistant", "content": "a1"},
]
out = apply_anthropic_cache_control_long_lived(
msgs, long_lived_ttl="1h", rolling_ttl="5m",
)
# Last user message: cache_control on the wrapped text part should be 5m
last = out[-1]
c = last["content"]
assert isinstance(c, list)
assert c[-1]["cache_control"] == {"type": "ephemeral"} # 5m has no ttl key
def test_string_system_falls_back_to_envelope_marker(self):
"""When the caller didn't split the system message, we still place a marker."""
msgs = [
{"role": "system", "content": "Single string system"},
{"role": "user", "content": "u1"},
]
out = apply_anthropic_cache_control_long_lived(msgs)
sys_content = out[0]["content"]
# Wrapped into a list and the (now sole) block gets the 1h marker
assert isinstance(sys_content, list)
assert sys_content[0]["cache_control"] == {"type": "ephemeral", "ttl": "1h"}
def test_does_not_mutate_input(self):
msgs = [
{"role": "system", "content": [{"type": "text", "text": "S"}]},
{"role": "user", "content": "u1"},
]
before = copy.deepcopy(msgs)
apply_anthropic_cache_control_long_lived(msgs)
assert msgs == before
def test_max_4_breakpoints_with_split_system(self):
msgs = [
{"role": "system", "content": [{"type": "text", "text": "S"}, {"type": "text", "text": "V"}]},
] + [
{"role": "user" if i % 2 == 0 else "assistant", "content": f"msg{i}"}
for i in range(10)
]
out = apply_anthropic_cache_control_long_lived(msgs)
count = 0
for m in out:
c = m.get("content")
if isinstance(c, list):
for item in c:
if isinstance(item, dict) and "cache_control" in item:
count += 1
elif "cache_control" in m:
count += 1
# 1 system block + last 2 messages = 3 breakpoints from this function.
# tools[-1] is marked separately (not via this function), so a 4th
# breakpoint can be added at API-call time.
assert count == 3
tests/agent/test_prompt_caching_live.py
-112
View File
@@ -1,112 +0,0 @@
"""Live E2E: long-lived prefix caching on Claude via OpenRouter.
Run only when LIVE_OR_KEY env var is set. Skipped under the normal hermetic
test suite (which unsets credentials).
"""
import os, sys, tempfile, time, shutil, pytest
# Probe for the key BEFORE conftest unsets it
_LIVE_KEY = os.environ.get("OPENROUTER_API_KEY") or os.environ.get("LIVE_OR_KEY")
if not _LIVE_KEY:
# Try to read directly from .env
env_path = os.path.expanduser("~/.hermes/.env")
if os.path.exists(env_path):
with open(env_path) as f:
for line in f:
if line.startswith("OPENROUTER_API_KEY="):
_LIVE_KEY = line.strip().split("=", 1)[1].strip().strip('"').strip("'")
break
pytestmark = pytest.mark.skipif(
not _LIVE_KEY,
reason="set OPENROUTER_API_KEY (or LIVE_OR_KEY) to run live cache test",
)
def test_long_lived_prefix_cache_e2e_openrouter(tmp_path, monkeypatch):
"""Two AIAgent runs in fresh sessions: call 1 writes cache, call 2 reads it."""
monkeypatch.setenv("HERMES_HOME", str(tmp_path))
# The hermetic conftest unsets OPENROUTER_API_KEY — restore for this test
monkeypatch.setenv("OPENROUTER_API_KEY", _LIVE_KEY)
# Minimal config — but with enough toolset/guidance to exceed Anthropic's
# ~1024-token minimum-cacheable-prefix threshold. Anthropic silently
# ignores cache_control markers on small blocks.
import yaml
cfg_path = tmp_path / "config.yaml"
cfg_path.write_text(yaml.safe_dump({
"model": {"provider": "openrouter", "default": "anthropic/claude-haiku-4.5"},
"prompt_caching": {"long_lived_prefix": True, "long_lived_ttl": "1h", "cache_ttl": "5m"},
"agent": {"tool_use_enforcement": True}, # adds substantial guidance text
"memory": {"provider": ""},
"compression": {"enabled": False},
}))
from run_agent import AIAgent
def make_agent():
return AIAgent(
api_key=_LIVE_KEY,
base_url="https://openrouter.ai/api/v1",
provider="openrouter",
model="anthropic/claude-haiku-4.5",
api_mode="chat_completions",
# Use the default toolset roster — the tools array (~13k tokens
# for ~35 tools) is what carries the bulk of the cross-session
# cache value. With a tiny toolset the cached prefix can fall
# below Anthropic Haiku's 2048-token minimum cacheable size and
# the marker is silently ignored.
enabled_toolsets=None,
quiet_mode=True,
skip_context_files=True,
skip_memory=True,
save_trajectories=False,
)
a1 = make_agent()
assert a1._use_prompt_caching is True, "policy should enable caching for Claude on OR"
assert a1._use_long_lived_prefix_cache is True, "long-lived path should activate"
parts = a1._build_system_prompt_parts()
print(f"\nstable={len(parts['stable']):,} ctx={len(parts['context']):,} volatile={len(parts['volatile']):,} chars")
print(f"tool count: {len(a1.tools or [])}")
# Use distinct user messages each call so OpenRouter's response cache
# doesn't short-circuit the upstream Anthropic call (we need real
# Anthropic billing visibility to verify cache_creation/cache_read).
USER_1 = "Reply with the single word ALPHA."
USER_2 = "Reply with the single word BRAVO."
print("\n--- Call 1 (cold) ---")
r1 = a1.run_conversation(USER_1, conversation_history=[])
print(f"final_response[:80]: {(r1.get('final_response') or '')[:80]!r}")
cr1 = a1.session_cache_read_tokens
cw1 = a1.session_cache_write_tokens
print(f"call1: cache_read={cr1} cache_write={cw1}")
# Wait so cache settles, then fresh agent (NEW SESSION) for cross-session read
time.sleep(2)
a2 = make_agent()
assert a2.session_id != a1.session_id, "second agent must have a new session"
print("\n--- Call 2 (warm, NEW session, different user msg) ---")
r2 = a2.run_conversation(USER_2, conversation_history=[])
print(f"final_response[:80]: {(r2.get('final_response') or '')[:80]!r}")
cr2 = a2.session_cache_read_tokens
cw2 = a2.session_cache_write_tokens
print(f"call2: cache_read={cr2} cache_write={cw2}")
print(f"\n=== VERDICT ===")
print(f" call1 wrote {cw1:,} cache tokens, read {cr1:,}")
print(f" call2 wrote {cw2:,} cache tokens, read {cr2:,}")
if cw1:
print(f" cross-session read fraction: cr2/cw1 = {cr2/cw1:.2%}")
# Assertions
assert cw1 > 0, f"call 1 must write cache (got {cw1}); long-lived layout not reaching wire"
assert cr2 > 0, (
f"call 2 must read cache cross-session (got {cr2}); "
f"stable prefix is not byte-stable across sessions"
)
assert cr2 >= 1000, f"cache_read on call 2 ({cr2}) too small to indicate real reuse"
tests/agent/transports/test_chat_completions.py
+2 -1
View File
@@ -147,11 +147,12 @@ class TestChatCompletionsBuildKwargs:
]
def test_nous_tags(self, transport):
from agent.portal_tags import nous_portal_tags
from providers import get_provider_profile
profile = get_provider_profile("nous")
msgs = [{"role": "user", "content": "Hi"}]
kw = transport.build_kwargs(model="gpt-4o", messages=msgs, provider_profile=profile)
assert kw["extra_body"]["tags"] == ["product=hermes-agent"]
assert kw["extra_body"]["tags"] == nous_portal_tags()
def test_reasoning_default(self, transport):
msgs = [{"role": "user", "content": "Hi"}]
tests/gateway/test_restart_drain.py
+2 -1
View File
@@ -7,6 +7,7 @@ from unittest.mock import AsyncMock, MagicMock
import pytest
import gateway.run as gateway_run
from agent.i18n import t
from gateway.platforms.base import MessageEvent, MessageType
from gateway.restart import DEFAULT_GATEWAY_RESTART_DRAIN_TIMEOUT
from gateway.session import SessionEntry, build_session_key
@@ -32,7 +33,7 @@ async def test_restart_command_while_busy_requests_drain_without_interrupt(monke
result = await runner._handle_message(event)
assert result == "⏳ Draining 1 active agent(s) before restart..."
assert result == t("gateway.draining", count=1)
running_agent.interrupt.assert_not_called()
runner.request_restart.assert_called_once_with(detached=True, via_service=False)
tests/providers/test_profile_wiring.py
+2 -1
View File
@@ -273,12 +273,13 @@ class TestRequestOverridesParity:
def test_extra_body_override_merges_with_provider_body(self, transport):
"""Override extra_body merges WITH provider extra_body, not replaces."""
from agent.portal_tags import nous_portal_tags
kw = transport.build_kwargs(
model="hermes-3", messages=_msgs(), tools=None,
provider_profile=get_provider_profile("nous"),
request_overrides={"extra_body": {"custom": True}},
)
assert kw["extra_body"]["tags"] == ["product=hermes-agent"] # from profile
assert kw["extra_body"]["tags"] == nous_portal_tags() # from profile
assert kw["extra_body"]["custom"] is True # from override
def test_top_level_override(self, transport):
tests/providers/test_provider_profiles.py
+2 -1
View File
@@ -210,9 +210,10 @@ class TestOpenRouterProfile:
class TestNousProfile:
def test_tags(self):
from agent.portal_tags import nous_portal_tags
p = get_provider_profile("nous")
body = p.build_extra_body()
assert body["tags"] == ["product=hermes-agent"]
assert body["tags"] == nous_portal_tags()
def test_auth_type(self):
p = get_provider_profile("nous")
tests/providers/test_transport_parity.py
+2 -1
View File
@@ -165,13 +165,14 @@ class TestNousParity:
"""Nous: product tags, reasoning, omit when disabled."""
def test_tags(self, transport):
from agent.portal_tags import nous_portal_tags
kw = transport.build_kwargs(
model="hermes-3-llama-3.1-405b",
messages=_simple_messages(),
tools=None,
provider_profile=get_provider_profile("nous"),
)
assert kw["extra_body"]["tags"] == ["product=hermes-agent"]
assert kw["extra_body"]["tags"] == nous_portal_tags()
def test_reasoning_omitted_when_disabled(self, transport):
"""Nous special case: reasoning omitted entirely when disabled."""
tests/run_agent/test_anthropic_prompt_cache_policy.py
-124
View File
@@ -330,127 +330,3 @@ class TestExplicitOverrides:
# Long-lived prefix cache policy (cross-session 1h tier)
# ─────────────────────────────────────────────────────────────────────
class TestSupportsLongLivedAnthropicCache:
"""Narrower than _anthropic_prompt_cache_policy — only Claude on the 4
explicitly-validated endpoints get the long-lived layout."""
def test_native_anthropic_claude_supported(self):
agent = _make_agent(
provider="anthropic",
base_url="https://api.anthropic.com",
api_mode="anthropic_messages",
model="claude-sonnet-4.6",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_anthropic_oauth_supported(self):
# OAuth uses the same transport as native Anthropic
agent = _make_agent(
provider="anthropic",
base_url="https://api.anthropic.com",
api_mode="anthropic_messages",
model="claude-opus-4.6",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_openrouter_claude_supported(self):
agent = _make_agent(
provider="openrouter",
base_url="https://openrouter.ai/api/v1",
api_mode="chat_completions",
model="anthropic/claude-sonnet-4.6",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_nous_portal_claude_supported(self):
# Nous Portal proxies to OpenRouter — same wire format
agent = _make_agent(
provider="nous",
base_url="https://inference-api.nousresearch.com/v1",
api_mode="chat_completions",
model="anthropic/claude-opus-4.7",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_nous_portal_qwen_supported(self):
# Portal Qwen rides the same OpenRouter-equivalent transport as
# Portal Claude; long-lived (1h cross-session) cache_control
# markers apply identically.
agent = _make_agent(
provider="nous",
base_url="https://inference-api.nousresearch.com/v1",
api_mode="chat_completions",
model="qwen3.6-plus",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_nous_portal_qwen_vendored_slug_supported(self):
agent = _make_agent(
provider="nous",
base_url="https://inference-api.nousresearch.com/v1",
api_mode="chat_completions",
model="qwen/qwen3.6-plus",
)
assert agent._supports_long_lived_anthropic_cache() is True
def test_nous_portal_non_claude_non_qwen_rejected(self):
# Portal long-lived cache scope mirrors policy: Claude or Qwen only.
agent = _make_agent(
provider="nous",
base_url="https://inference-api.nousresearch.com/v1",
api_mode="chat_completions",
model="openai/gpt-5.4",
)
assert agent._supports_long_lived_anthropic_cache() is False
def test_openrouter_non_claude_rejected(self):
agent = _make_agent(
provider="openrouter",
base_url="https://openrouter.ai/api/v1",
api_mode="chat_completions",
model="openai/gpt-5.4",
)
assert agent._supports_long_lived_anthropic_cache() is False
def test_third_party_anthropic_gateway_rejected(self):
# MiniMax / Kimi / etc. — anthropic-wire but not in our validated list
agent = _make_agent(
provider="minimax",
base_url="https://api.minimax.io/anthropic",
api_mode="anthropic_messages",
model="minimax-m2.7",
)
assert agent._supports_long_lived_anthropic_cache() is False
def test_alibaba_dashscope_rejected(self):
agent = _make_agent(
provider="alibaba",
base_url="https://dashscope.aliyuncs.com/api/v1/anthropic",
api_mode="anthropic_messages",
model="qwen3.5-plus",
)
assert agent._supports_long_lived_anthropic_cache() is False
def test_opencode_qwen_rejected(self):
agent = _make_agent(
provider="opencode-go",
base_url="https://api.opencode-go.example/v1",
api_mode="chat_completions",
model="qwen3.6-plus",
)
assert agent._supports_long_lived_anthropic_cache() is False
def test_fallback_target_evaluated_independently(self):
# Starting on a non-supported provider, falling back to OpenRouter Claude
agent = _make_agent(
provider="minimax",
base_url="https://api.minimax.io/anthropic",
api_mode="anthropic_messages",
model="minimax-m2.7",
)
assert agent._supports_long_lived_anthropic_cache(
provider="openrouter",
base_url="https://openrouter.ai/api/v1",
api_mode="chat_completions",
model="anthropic/claude-sonnet-4.6",
) is True
tests/run_agent/test_provider_parity.py
+2 -1
View File
@@ -343,11 +343,12 @@ class TestBuildApiKwargsAIGateway:
class TestBuildApiKwargsNousPortal:
def test_includes_nous_product_tags(self, monkeypatch):
from agent.portal_tags import nous_portal_tags
agent = _make_agent(monkeypatch, "nous", base_url="https://inference-api.nousresearch.com/v1")
messages = [{"role": "user", "content": "hi"}]
kwargs = agent._build_api_kwargs(messages)
extra = kwargs.get("extra_body", {})
assert extra.get("tags") == ["product=hermes-agent"]
assert extra.get("tags") == nous_portal_tags()
def test_uses_chat_completions_format(self, monkeypatch):
agent = _make_agent(monkeypatch, "nous", base_url="https://inference-api.nousresearch.com/v1")
tests/test_ctx_halving_fix.py
-1
View File
@@ -169,7 +169,6 @@ class TestEphemeralMaxOutputTokens:
agent.reasoning_config = None
agent._is_anthropic_oauth = False
agent._ephemeral_max_output_tokens = None
agent._use_long_lived_prefix_cache = False
compressor = MagicMock()
compressor.context_length = 200_000