mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-27 17:58:07 +00:00
feat(gateway): add /context command for a detailed context-window view
A dedicated /context (alias /ctx) gateway slash command that gives a full
context-window view with:
- Usage gauge: visual bar + fraction + percentage + headroom
- Auto-compression threshold and how far away it is
- Compression count and how much the last one freed
- Cumulative session throughput (explicitly labelled as throughput,
NOT context size — each call re-sends the window)
- Cascading fallback: running agent → cached agent → SessionStore metadata
→ rough transcript estimate
Not included (per current-main design):
- Cache reporting removed: commit 446b8e239 intentionally removed cache
reporting from user-facing surfaces because providers that omit cached-token
details produce misleading values
- Sync DB calls replaced with async_session_store (current main requires
AsyncSessionStore with await)
Also rewords the /status tokens line from 'Cumulative API tokens (re-sent
each call)' to 'Lifetime tokens billed: ... (not your current context size;
use /context)' to reduce the recurring confusion that the cumulative figure
is the current context window.
Fixes salvation of PR #52184 (salvage commit replaces a 12K-commit-behind
fork branch with a fresh implementation against current main, incorporating
reviewer feedback from @whoislikemiha and the hermes-sweeper).
This commit is contained in:
CharlesMcquade
Teknium
parent
3dccc45032
commit
4fe2fecf54
20 changed files with 654 additions and 22 deletions
|
|
@ -100,7 +100,7 @@ async def test_status_command_reports_running_agent_without_interrupt(monkeypatc
|
|||
result = await runner._handle_message(_make_event("/status"))
|
||||
|
||||
assert "**Session ID:** `sess-1`" in result
|
||||
assert "**Cumulative API tokens (re-sent each call):** 321" in result
|
||||
assert "**Lifetime tokens billed:** 321" in result
|
||||
assert "**Agent Running:** Yes ⚡" in result
|
||||
assert "**Title:**" not in result
|
||||
running_agent.interrupt.assert_not_called()
|
||||
|
|
@ -153,7 +153,7 @@ async def test_status_command_reads_token_totals_from_session_db():
|
|||
result = await runner._handle_message(_make_event("/status"))
|
||||
|
||||
# 1000 + 250 + 500 + 100 + 50 = 1,900
|
||||
assert "**Cumulative API tokens (re-sent each call):** 1,900" in result
|
||||
assert "**Lifetime tokens billed:** 1,900" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
|
|
@ -174,7 +174,7 @@ async def test_status_command_tokens_zero_when_session_db_row_missing():
|
|||
|
||||
result = await runner._handle_message(_make_event("/status"))
|
||||
|
||||
assert "**Cumulative API tokens (re-sent each call):** 0" in result
|
||||
assert "**Lifetime tokens billed:** 0" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
|
|
@ -212,8 +212,7 @@ async def test_status_command_includes_live_agent_model_and_context():
|
|||
|
||||
assert "**Model:** `openai/gpt-test` (openai)" in result
|
||||
assert "**Context:** 12,345 / 100,000 (12%)" in result
|
||||
assert "**Cumulative API tokens (re-sent each call):** 1,250" in result
|
||||
assert "1,250 (cumulative)" not in result
|
||||
assert "**Lifetime tokens billed:** 1,250" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
|
|
@ -245,7 +244,7 @@ async def test_status_command_includes_persisted_model_and_context_when_agent_no
|
|||
|
||||
assert "**Model:** `openai/gpt-persisted` (openai-codex)" in result
|
||||
assert "**Context:** 24,000 / 272,000 (9%)" in result
|
||||
assert "**Cumulative API tokens (re-sent each call):** 2,500" in result
|
||||
assert "**Lifetime tokens billed:** 2,500" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
|
|
@ -821,3 +820,152 @@ async def test_post_delivery_callback_generation_snapshot_happens_after_bind():
|
|||
assert fired == []
|
||||
assert session_key in adapter._post_delivery_callbacks
|
||||
assert adapter._post_delivery_callbacks[session_key][0] == 2
|
||||
|
||||
|
||||
# ── /context command tests ────────────────────────────────────────────────
|
||||
|
||||
def _stub_agent(**overrides) -> SimpleNamespace:
|
||||
"""Build a stub agent with the attributes _handle_context_command reads."""
|
||||
props = dict(
|
||||
model="openai/gpt-test",
|
||||
context_compressor=SimpleNamespace(
|
||||
last_prompt_tokens=47_231,
|
||||
context_length=200_000,
|
||||
threshold_tokens=100_000,
|
||||
threshold_percent=0.5,
|
||||
compression_count=2,
|
||||
_last_compression_savings_pct=63.0,
|
||||
),
|
||||
session_api_calls=47,
|
||||
session_input_tokens=410_000,
|
||||
session_output_tokens=38_000,
|
||||
session_reasoning_tokens=12_000,
|
||||
session_total_tokens=3_158_641,
|
||||
session_cache_read_tokens=2_900_000,
|
||||
session_cache_write_tokens=48_000,
|
||||
)
|
||||
props.update(overrides)
|
||||
return SimpleNamespace(**props)
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_context_command_live_agent():
|
||||
"""/context with a live running agent shows the full view: gauge,
|
||||
compression, and throughput — but NOT cache stats."""
|
||||
session_entry = SessionEntry(
|
||||
session_key=build_session_key(_make_source()),
|
||||
session_id="sess-1",
|
||||
created_at=datetime.now(),
|
||||
updated_at=datetime.now(),
|
||||
platform=Platform.TELEGRAM,
|
||||
chat_type="dm",
|
||||
)
|
||||
runner = _make_runner(session_entry)
|
||||
session_key = session_entry.session_key
|
||||
agent = _stub_agent()
|
||||
runner._running_agents[session_key] = agent
|
||||
|
||||
result = await runner._handle_context_command(_make_event("/context"))
|
||||
|
||||
assert "🧠 **Context Window**" in result
|
||||
assert "Model: `openai/gpt-test`" in result
|
||||
assert "Window: 200,000 tokens" in result
|
||||
assert "In use: 47,231 / 200,000 (24%)" in result
|
||||
assert "Headroom to limit: 152,769 tokens" in result
|
||||
# Compression section
|
||||
assert "Auto-compresses at: 100,000 (50%)" in result
|
||||
assert "Compressions this session: 2" in result
|
||||
assert "Last compression freed: 63% of context" in result
|
||||
# Throughput — NOT cache
|
||||
assert "Session totals (cumulative across 47 API calls)" in result
|
||||
assert "Input 410,000" in result
|
||||
assert "Output 38,000" in result
|
||||
assert "Reasoning 12,000" in result
|
||||
assert "Total billed: 3,158,641" in result
|
||||
assert "each call re-sends the window above" in result
|
||||
# Cache stats must NOT appear (removed per design)
|
||||
assert "Cache read" not in result
|
||||
assert "Cache write" not in result
|
||||
assert "Cache hit" not in result
|
||||
assert "Hit rate" not in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_context_command_over_threshold():
|
||||
"""When used >= threshold, the over-threshold warning is shown."""
|
||||
session_entry = SessionEntry(
|
||||
session_key=build_session_key(_make_source()),
|
||||
session_id="sess-2",
|
||||
created_at=datetime.now(),
|
||||
updated_at=datetime.now(),
|
||||
platform=Platform.TELEGRAM,
|
||||
chat_type="dm",
|
||||
)
|
||||
runner = _make_runner(session_entry)
|
||||
session_key = session_entry.session_key
|
||||
agent = _stub_agent(
|
||||
context_compressor=SimpleNamespace(
|
||||
last_prompt_tokens=150_000,
|
||||
context_length=200_000,
|
||||
threshold_tokens=100_000,
|
||||
threshold_percent=0.5,
|
||||
compression_count=5,
|
||||
_last_compression_savings_pct=40.0,
|
||||
)
|
||||
)
|
||||
runner._running_agents[session_key] = agent
|
||||
|
||||
result = await runner._handle_context_command(_make_event("/context"))
|
||||
|
||||
assert "⚠️" in result
|
||||
assert "Over auto-compression threshold" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_context_command_no_agent_transcript_fallback():
|
||||
"""When no agent is resident and session_entry has no last_prompt_tokens,
|
||||
/context falls back to a transcript estimate."""
|
||||
session_entry = SessionEntry(
|
||||
session_key=build_session_key(_make_source()),
|
||||
session_id="sess-3",
|
||||
created_at=datetime.now(),
|
||||
updated_at=datetime.now(),
|
||||
platform=Platform.TELEGRAM,
|
||||
chat_type="dm",
|
||||
last_prompt_tokens=0, # No live context data
|
||||
)
|
||||
runner = _make_runner(session_entry)
|
||||
# Stub the transcript so estimate_messages_tokens_rough has something to work with
|
||||
runner.session_store.load_transcript.return_value = [
|
||||
{"role": "user", "content": "Hello"},
|
||||
{"role": "assistant", "content": "Hi there!"},
|
||||
{"role": "user", "content": "What's my balance?"},
|
||||
{"role": "assistant", "content": "Your balance is $1,000."},
|
||||
]
|
||||
|
||||
result = await runner._handle_context_command(_make_event("/context"))
|
||||
|
||||
assert "🧠 **Context Window**" in result
|
||||
assert "Estimated context:" in result
|
||||
assert "4 messages" in result
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_context_command_no_data():
|
||||
"""When there's no agent, no session data, and no transcript,
|
||||
/context returns the no-data message."""
|
||||
session_entry = SessionEntry(
|
||||
session_key=build_session_key(_make_source()),
|
||||
session_id="sess-4",
|
||||
created_at=datetime.now(),
|
||||
updated_at=datetime.now(),
|
||||
platform=Platform.TELEGRAM,
|
||||
chat_type="dm",
|
||||
last_prompt_tokens=0,
|
||||
)
|
||||
runner = _make_runner(session_entry)
|
||||
runner.session_store.load_transcript.return_value = []
|
||||
|
||||
result = await runner._handle_context_command(_make_event("/context"))
|
||||
|
||||
assert "No context data available yet" in result
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue