docs(memory): document memgw Memory Gateway provider

[cursor]

What does this PR do?

Documents the bundled memgw Memory Gateway provider across the Hermes user, reference, and contributor docs. This closes drift from the recent provider implementation where only the plugin README described setup and behavior while the website still listed the older provider set.

Related Issue

N/A

Type of Change

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

Changes Made

• Added a full memgw (Memory Gateway) section to website/docs/user-guide/features/memory-providers.md covering setup, auth modes, $HERMES_HOME/memgw.json, automatic lifecycle behavior, circuit breaker behavior, and backend links.
• Updated provider inventories in memory.md, overview.md, integrations/index.md, and reference/cli-commands.md from 8 to 9 providers.
• Added MEMGW_API_URL and MEMGW_API_KEY to the environment variable reference.
• Added memgw_recall, memgw_retain, and memgw_reflect notes to the tools reference.
• Updated developer docs for memory-provider hooks and gateway lifecycle/user scoping (queue_prefetch, sync_turn, on_session_switch, on_delegation).
• Updated AGENTS.md, CONTRIBUTING.md, and plugins/memory/memgw/README.md to align the in-tree provider policy with the fork-specific memgw exception and current provider behavior.

Codepaths Covered

• plugins/memory/memgw/__init__.py: provider availability, config loading, loopback auth rules, prefetch/session switch, async turn sync, delegation recording, tool handling, shutdown joins.
• plugins/memory/memgw/client.py: Streamable-HTTP MCP endpoint normalization, bearer auth, timeout cancellation, and blocking wrapper over async MCP calls.
• Gateway memory-provider flow documented in website/docs/developer-guide/gateway-internals.md.

Key Knowledge Gaps Addressed

• Website/reference docs did not mention memgw as an available memory provider.
• Users lacked setup guidance for cloud vs exact-loopback local Memory Gateway auth.
• Operators lacked docs for recall_limit, prefetch_method, circuit breaker behavior, session-switch cache invalidation, and gateway user_id scoping.
• Contributor policy docs listed the closed provider set but did not include or explain the fork-specific memgw exception.

How to Test

1. git show --check --stat --oneline HEAD
2. rg "8 external|Eight providers|8 providers|Available providers: honcho|\\(honcho, mem0, supermemory|providers include \\*\\*honcho, mem0" /agent/repos/hermes-agent --glob '*.md'
3. cd website && npm ci && npm run build

Notes: the Docusaurus build exits successfully and generates static files. It still reports pre-existing warning-only broken links/anchors in unrelated docs/locales (for example rl-training.md, tool-gateway.md, and Korean localized doc links); none reference the new memgw docs.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for stale provider-list references in the docs
• My PR contains only changes related to this documentation update
• I've run pytest tests/ -q and all tests pass (N/A: docs-only)
• I've added tests for my changes (N/A: docs-only)
• I've tested on my platform: Cursor Cloud Linux

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 (N/A: documented existing env/config only)
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact — or N/A (N/A: docs-only)
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A (N/A: no behavior change)

Screenshots / Logs

Docusaurus build completed successfully with existing unrelated broken-link/anchor warnings.

  

[dizhaky]

Closing: original commit was unsigned (cursor-agent). Replacing with signed commit in #42.