Skip to content

perf(gateway): reduce Telegram end-to-end response latency#10388

Closed
wilsen0 wants to merge 3 commits into
NousResearch:mainfrom
wilsen0:perf/gateway-telegram-latency
Closed

perf(gateway): reduce Telegram end-to-end response latency#10388
wilsen0 wants to merge 3 commits into
NousResearch:mainfrom
wilsen0:perf/gateway-telegram-latency

Conversation

@wilsen0

@wilsen0 wilsen0 commented Apr 15, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Lower Telegram text-batch ingress delay (0.6s → 0.3s default, 2.0s → 1.0s for near-4096 splits) with adaptive fast-path for short messages (≤320 chars: 0.18s, ≤1024 chars: 0.24s), cutting the fixed "wait-for-more-chunks" window by ~50%.
  • Tune streaming edit rhythm for faster first visible token: edit_interval 1.0s → 0.8s, buffer_threshold 40 → 24 chars — balanced against Telegram's ~1 edit/s flood limit.
  • Switch Telegram default tool_progress from "all" to "new" to reduce edit pressure and flood-control risk.
  • Add bounded env-var parser _env_float_clamped() with math.isfinite() guard to reject NaN/Inf at the source.
  • Unify streaming defaults to a single source of truth (DEFAULT_STREAMING_EDIT_INTERVAL / DEFAULT_STREAMING_BUFFER_THRESHOLD / DEFAULT_STREAMING_CURSOR in gateway/config.py); StreamConsumerConfig imports from there, eliminating dual-definition drift risk.
  • Extract _is_gateway_streaming_enabled() helper for consistent platform-streaming resolution across both gateway code paths, with proper True/False/None override semantics.
  • Fix already_sent dedup logic to account for response_previewed and gate stream-consumer already_sent behind effective streaming state, preventing false suppression of final responses.

Test plan

  • tests/gateway/test_display_config.py — platform default assertions updated for Telegram tool_progress: "new"
  • tests/gateway/test_proxy_mode.py — added TestGatewayStreamingResolver unit tests (5 cases)
  • tests/gateway/test_run_progress_topics.py — verified no regression in already_sent / previewed flow
  • tests/gateway/test_text_batching.py — adaptive delay logic unchanged for non-Telegram platforms
  • tests/gateway/test_stream_consumer.py — streaming consumer defaults now flow from unified constants
  • tests/gateway/test_duplicate_reply_suppression.py — dedup logic verified
  • 155 gateway tests passed, 0 failures

teknium1 added 3 commits April 15, 2026 20:54
@teknium1
fix(anthropic): preserve third-party thinking continuity
d2f043f 
Downgrade third-party thinking blocks to text so reasoning context survives across turns while removing redacted payloads and stale signatures. Add regression tests for third-party thinking conversion and keep z.ai preserved-thinking behavior server-driven by removing explicit clear_thinking injection.
@teknium1
perf(gateway): reduce telegram response latency while preserving deli…
df7d638 
…very stability
@teknium1
fix(gateway): harden env float parsing and unify streaming defaults t…
c8361c1 
…o single source
@alt-glitch alt-glitch added type/perf Performance improvement or optimization P2 Medium — degraded but workaround exists comp/gateway Gateway runner, session dispatch, delivery platform/telegram Telegram bot adapter labels Apr 26, 2026
@teknium1 teknium1 mentioned this pull request May 11, 2026
Merged
teknium1 pushed a commit that referenced this pull request May 11, 2026
@wilsen0 @teknium1
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
ac95b8c 
…eplies

Re-authored against current main from PR #10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes #10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
@teknium1

teknium1 commented May 11, 2026

Copy link
Copy Markdown
Contributor

Merged via salvage PR #23587 (#23587). Your authorship preserved on the substantive commit.

Re-authored against current main rather than cherry-picked — your branch was 3800+ commits behind and would have reverted unrelated work (Slack tool_progress override #14663, Platform plugin discovery, MSGRAPH_WEBHOOK + YUANBAO members). The salvage carries only the perf intent forward:

  • Adaptive text-batch fast-path tiers (≤320 cp: ~180ms; ≤1024 cp: ~240ms; else: configured cap, with min() composition so tightening the cap wins on every tier)
  • Lowered defaults: ingress 0.6→0.3s, split 2.0→1.0s, edit_interval 1.0→0.8s, buffer_threshold 40→24
  • _env_float_clamped helper rejects NaN/Inf and applies bounds
  • Telegram tool_progress 'all'→'new' (Slack 'off' override preserved)
  • Single source of truth for streaming defaults — fixes the prior gateway/config.py vs stream_consumer.py drift

Salvage dropped your unrelated d2f043f (anthropic thinking) commit and a boot_md.py builtin gateway hook that wasn't in scope. Added 11 regression tests for the env helper + tier composition. Composes cleanly with the native draft transport (#23512). Thanks for the perf work — short Telegram replies feel noticeably snappier now.

rmulligan pushed a commit to rmulligan/hermes-agent that referenced this pull request May 11, 2026
@wilsen0
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
7cd18f3 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
JinyuID pushed a commit to JinyuID/hermes-agent that referenced this pull request May 11, 2026
@wilsen0 @teknium1
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
e384b35 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
mxdhavgautam pushed a commit to mxdhavgautam/hermes-agent that referenced this pull request May 13, 2026
@wilsen0
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
c999996 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
(cherry picked from commit ac95b8c)
02356abc pushed a commit to 02356abc/hermes-agent that referenced this pull request May 14, 2026
@wilsen0 @teknium1
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
ae17a70 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
jsboige pushed a commit to jsboige/hermes-agent that referenced this pull request May 14, 2026
@wilsen0 @teknium1
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
b0d14c4 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
AlexFoxD pushed a commit to AlexFoxD/hermes-agent that referenced this pull request May 21, 2026
@wilsen0 @AlexFoxD
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
3e5d8cf 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
gweeteve pushed a commit to gweeteve/hermes-agent that referenced this pull request Jun 2, 2026
@wilsen0 @teknium1
perf(gateway): tune Telegram cadence + adaptive fast-path for short r…
c6d2106 
…eplies

Re-authored against current main from PR NousResearch#10388 by @wilsen0.  The
original branch is 3800+ commits stale and could not be cherry-picked
without reverting unrelated work; this change carries only the perf
intent forward.

Tuning summary
==============

Text-batch ingress (gateway/platforms/telegram.py):
  - HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS default 0.6 -> 0.3
  - HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS default 2.0 -> 1.0
  - Adaptive fast-path tiers in _flush_text_batch:
      total <= 320 cp -> min(cap, 0.18)
      total <= 1024 cp -> min(cap, 0.24)
      else            -> cap
    A single short reply now reaches the agent in ~180ms instead of
    600ms.  Tier constants compose with the configured cap via min()
    so an operator who tightens HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS
    below 0.18 still wins on every tier.
  - _env_float_clamped helper replaces bare float(os.getenv()).
    Rejects NaN / Inf, applies optional min/max bounds.  Used for
    text-batch + media-batch knobs.  Prevents asyncio.sleep(NaN)
    crashes when an operator typos an env var.

Stream cadence (gateway/config.py + stream_consumer.py):
  - StreamingConfig.edit_interval default 1.0s -> 0.8s
  - StreamingConfig.buffer_threshold default 40 -> 24 chars
  - DEFAULT_STREAMING_EDIT_INTERVAL / BUFFER_THRESHOLD / CURSOR are now
    a single source of truth.  StreamConsumerConfig imports them
    instead of duplicating the literals; the prior dual-source drift
    is fixed.

Tool progress (gateway/display_config.py):
  - Telegram default tool_progress 'all' -> 'new'.  Inside
    Telegram's ~1 edit/s flood envelope the 'all' default would
    accumulate edit pressure on busy chats; 'new' shows only the
    leading bubble per tool batch and feels less spammy.
  - Slack tier_low override (tool_progress='off') is preserved.

Composition with native draft streaming (NousResearch#23512)
================================================

The mid-stream cadence (edit_interval, buffer_threshold) gates BOTH
the draft path (send_draft) and the edit path (edit_message), so the
tighter cadence helps native draft as much as edit-based.  The
text-batch fast-path applies before the consumer starts, so it speeds
up the first-token latency on every transport.  No conflict.

Stale-base avoidance
====================

Re-authored from scratch rather than cherry-picked.  Dropped from the
original branch:
  - Unrelated d2f043f 'fix(anthropic): preserve third-party thinking
    continuity' commit
  - boot_md.py builtin gateway hook (unrelated)
  - Reverted Slack tool_progress='off' (NousResearch#14663) restoration
  - Reverted Platform plugin discovery, MSGRAPH_WEBHOOK, YUANBAO
    members deletion
  - 2300+ lines of run.py base-skew noise

Tests
=====

New tests/gateway/test_telegram_text_batch_perf.py:
  - 7 tests for _env_float_clamped (NaN, Inf, garbage, bounds).
  - 4 tests for the adaptive-tier composition rules.

Updated tests/gateway/test_display_config.py:
  - test_platform_default_when_no_user_config: 'all' -> 'new' for
    Telegram, with comment.
  - test_high_tier_platforms: split into Telegram-overrides-to-new
    and Discord-stays-all assertions.

Closes NousResearch#10388.

Co-authored-by: wilsen0 <132184373+wilsen0@users.noreply.github.com>
@rdnot rdnot mentioned this pull request Jun 12, 2026
Open
1 task
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

comp/gateway Gateway runner, session dispatch, delivery P2 Medium — degraded but workaround exists platform/telegram Telegram bot adapter type/perf Performance improvement or optimization

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@wilsen0 @teknium1 @alt-glitch