Issue Station
Sign In
Tools hereToolbox

hermes - 💡(How to fix) Fix Hardcoded platform timeouts break local model workflows — need unified timeout configuration

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…

Code Example

# Per-platform HTTP timeouts (new)
HERMES_MATTERMOST_HTTP_TIMEOUT=120
HERMES_SIGNAL_HTTP_TIMEOUT=120
HERMES_WHATSAPP_HTTP_TIMEOUT=60
HERMES_DISCORD_QUEUE_TIMEOUT=600

# Approval timeout (new)
HERMES_APPROVAL_TIMEOUT=600

# Retry backoff tuning (new)
HERMES_RETRY_BASE_DELAY=10
HERMES_RETRY_MAX_DELAY=300
RAW_BUFFERClick to expand / collapse

Problem

Users running local models (Ollama, llama.cpp, vLLM) frequently hit timeout errors at multiple levels of the Hermes stack. While agent-level timeouts are well-configured (auto-detection of local endpoints, env var overrides), platform adapter timeouts are hardcoded and cannot be tuned without code changes.

What's configurable (good):

  • HERMES_API_TIMEOUT (default 30min) — main LLM call
  • HERMES_STREAM_READ_TIMEOUT (default 120s, auto-raised for local endpoints)
  • HERMES_STREAM_STALE_TIMEOUT (auto-disabled for local endpoints)
  • HERMES_AGENT_TIMEOUT (inactivity-based, not wall-clock)
  • terminal.timeout, browser.command_timeout, auxiliary.*.timeout

What's hardcoded (breaks local model workflows):

ComponentTimeoutFileImpact
Mattermost HTTP30s GET / 60s POSTgateway/platforms/mattermost.py:117,136Message send fails if model response takes >30s to format
Signal HTTP30sgateway/platforms/signal.py:197Same
WhatsApp polling2-30sgateway/platforms/whatsapp.py:42-96Connection drops during long inference
Telegram DoH4sgateway/platforms/telegram_network.py:27Minor but compounds
Discord queues120-300sgateway/platforms/discord.py:2560-2739Voice/media timeout during slow generation
User approval300sgateway/run.py:6212Hard 5min wall-clock for approval decisions
Retry backoffbase=5s, max=120sagent/retry_utils.py:19-57OOM recovery on local models needs longer backoff

Use case

Running sparse Qwen 3 models locally — individual responses can take 2-5 minutes. The agent-level timeout is fine (30min), but platform adapters time out trying to send the response once it's ready. The retry backoff is also too aggressive for OOM recovery — local models crash and need 30-60s to restart, but Hermes retries after 5s and hits the same OOM.

Proposed fix

Expose hardcoded platform timeouts as env vars following the existing pattern:

# Per-platform HTTP timeouts (new)
HERMES_MATTERMOST_HTTP_TIMEOUT=120
HERMES_SIGNAL_HTTP_TIMEOUT=120
HERMES_WHATSAPP_HTTP_TIMEOUT=60
HERMES_DISCORD_QUEUE_TIMEOUT=600

# Approval timeout (new)
HERMES_APPROVAL_TIMEOUT=600

# Retry backoff tuning (new)
HERMES_RETRY_BASE_DELAY=10
HERMES_RETRY_MAX_DELAY=300

This follows the established HERMES_TELEGRAM_HTTP_* pattern (Telegram already has configurable HTTP timeouts via env vars).

Related issues

  • #5450 (per-request wall-clock timeout budget)
  • #12619 (command-type-aware terminal timeouts)
  • #16723 (terminal.timeout config mismatch)

Environment

  • Hermes version: 0.8.0
  • Models: Ollama (Qwen 3 sparse, qwen3.5:9b), local vLLM
  • Platforms: Telegram, Mattermost, Discord
  • OS: macOS + Linux (Docker)

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…
#api#GPU setup#container setup#orchestration issue#cache issue

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING