What does this PR do?

This PR improves Hermes gateway reliability during restarts and makes local speech-to-text more stable on WSL-style environments.

Specifically it:

• saves in-flight gateway turns when restart drain times out
• resumes those interrupted turns automatically after the gateway starts again
• switches local faster-whisper loading to CPU/int8 for reliability when partial CUDA setups cause libcublas failures
• updates the FAQ to recommend idempotent WSL startup wrappers so healthy gateways are not bounced by repeated startup triggers

Related Issue

Fixes #

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• 📝 Documentation update
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

gateway/run.py

• track active inbound events per session
• persist interrupted in-flight turns to a restart-resume file when restart drain times out
• replay those saved turns automatically after gateway startup

tools/transcription_tools.py

• force local faster-whisper to load with device="cpu" and compute_type="int8"
• avoids WSL/CTranslate2/CUDA partial-runtime failures such as missing libcublas.so.12

website/docs/reference/faq.md

• document idempotent WSL startup guidance
• warn against startup wrappers that restart an already healthy gateway

How to Test

1. Restart-resume behavior

• start Hermes gateway
• send a Telegram/gateway message that takes long enough to still be running during restart
• trigger a gateway restart while the task is active
• verify the gateway stores the interrupted turn and resumes it automatically after startup
• verify the user eventually gets a final reply instead of being left with only tool-progress output

2. STT stability

• send a voice note in a WSL environment with local STT enabled
• verify local transcription works in CPU/int8 mode
• verify the previous libcublas/CUDA failure path no longer occurs

3. Regression check

• ensure normal gateway requests still complete correctly
• ensure startup without interrupted turns does not attempt any replay

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform:
  • WSL2 / Linux gateway environment

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

Validation run locally:

• python3 -m py_compile gateway/run.py
• ./venv/bin/python -m pytest tests/gateway/test_session_boundary_hooks.py -q

Observed user-facing failure before fix:

• gateway could emit tool-progress messages on Telegram
• gateway restart/drain timeout could interrupt active work
• user would receive no final reply

Observed behavior after local fix:

• restart-resume logic compiles cleanly
• targeted gateway tests pass
• FAQ updated with idempotent WSL startup guidance

Notes

This PR intentionally excludes local environment-specific startup wrapper changes outside the repository.