feat(memory): Graphnosis local memory provider + MCP catalog entry

[nehloo]

Summary

Adds Graphnosis as a local encrypted memory provider and MCP catalog entry for Hermes.

Path	What users get
Memory provider	Auto-prefetch; graphnosis_recall, graphnosis_remember, graphnosis_stats
MCP catalog	hermes mcp install graphnosis → full mcp_graphnosis_* tool surface

Both connect to the Graphnosis desktop app via ~/.graphnosis/mcp.sock using the published npm relay @graphnosis/mcp-relay (Apache-2.0). No API key; cortex must be unlocked.

Reference bundle: https://github.com/nehloo-interactive/graphnosis-app/tree/main/integrations/hermes-agent

Changes

• plugins/memory/graphnosis/ — provider plugin + CLI (hermes graphnosis status, etc.)
• optional-mcps/graphnosis/manifest.yaml — catalog entry
• hermes_cli/mcp_config.py — graphnosis preset
• tests/agent/test_graphnosis_memory_provider.py
• Docs: integrations/graphnosis.md, Graphnosis section in memory-providers.md, sidebar entry

Test plan

• pytest tests/agent/test_graphnosis_memory_provider.py -q — 5 passed
• Graphnosis unlocked → hermes memory setup lists graphnosis
• hermes mcp catalog | grep -i graphnosis
• hermes mcp install graphnosis → new session exposes mcp_graphnosis_* tools
• hermes graphnosis status when provider active

Notes

Graphnosis also ships a Get Connected → Hermes wizard that writes the same ~/.hermes/config.yaml shape; this PR makes setup discoverable via native Hermes flows.

Maintained by Nehloo Interactive LLC (nehloo-interactive/graphnosis-app).

[nehloo]

Validation so far (Nehloo Interactive / Graphnosis)

• Graphnosis 1.22.0 on Mac Mini agent account — cortex unlock works (sidecar package.json fix)
• Hermes Desktop + local Llama; Graphnosis visible in setup
• pytest tests/agent/test_graphnosis_memory_provider.py -q — 5 passed
• MCP relay published: @graphnosis/mcp-relay@1.0.2+ on npm
• Integration source: https://github.com/nehloo-interactive/graphnosis-app/tree/main/integrations/hermes-agent

Happy to run any extra checks maintainers want (live hermes memory setup, hermes mcp install graphnosis, etc.).

[tonydwb]

Code Review Summary

Verdict: Approved

Overview

New memory provider plugin: Graphnosis local memory + MCP catalog entry. Implements a local memory system with MCP server for external agent integration.

Analysis

• 11 files all scoped to the new memory provider (plugin, CLI, client, MCP manifest, docs, sidebar)
• Well-structured plugin architecture: __init__.py, cli.py, client.py, plugin.yaml, README.md
• MCP catalog entry allows use with Claude Code, Codex, Gemini CLI
• Includes dedicated test file: tests/agent/test_graphnosis_memory_provider.py
• Documentation added: website/docs/user-guide/integrations/graphnosis.md and memory-providers.md updated

Looks Good

• Clean, well-scoped new feature
• Follows existing plugin architecture patterns
• Good test coverage
• Good documentation

---

Reviewed by Hermes Agent

[nehloo]

Companion Hermes Desktop UI PR is up: fathah/hermes-desktop#763

Adds Memory provider copy + download link, Capabilities → MCP install promo (hermes mcp install graphnosis), Discover bundled MCP entry (when optional-mcps/graphnosis is present), and Welcome onboarding card.

Depends on this agent PR for the manifest + memory plugin. Local validation: npm run typecheck + preload API surface tests pass on Desktop branch.

[tonydwb]

Code Review Summary

Verdict: Approved

New Graphnosis local memory provider: a file-based memory store with MCP (Model Context Protocol) catalog entry, local document ingestion, and session management. 11 files, +767/-4.

Changes

• plugins/memory/graphnosis/__init__.py, client.py, cli.py: Core Graphnosis implementation
• plugins/memory/graphnosis/README.md, plugin.yaml: Documentation and plugin manifest
• tests/agent/test_graphnosis_memory_provider.py: +75 lines test coverage
• Documentation updates for memory-providers and integrations

Quality

• 11 files — just below the 15-file high-surface-area threshold
• Good test coverage for core functionality
• No security concerns

---

Reviewed by Hermes Agent

[nehloo]

Hi — gentle nudge on this PR when you have a moment.

It adds Graphnosis as a local encrypted memory provider (graphnosis_recall / graphnosis_remember / graphnosis_stats) plus an MCP catalog entry (hermes mcp install graphnosis) via @graphnosis/mcp-relay and ~/.graphnosis/mcp.sock — no API key, cortex must be unlocked.

Validated: pytest tests/agent/test_graphnosis_memory_provider.py (5 passed), Graphnosis 1.22+/1.23.x, npm relay published. Companion Desktop UI PR: fathah/hermes-desktop#763 (rebased onto current main, now mergeable).

Two bot reviews are in; branch shows mergeable but blocked pending a human maintainer. Is anything else needed before merge, or could someone from the Nous team take a look? Happy to run live checks (hermes memory setup, hermes mcp install graphnosis) if helpful.

[teknium1]

Thanks for the substantial Graphnosis integration work, including the MCP catalog manifest, setup documentation, and the companion Desktop work referenced in the PR discussion.

This automated hermes-sweeper review is closing it under the standing memory-provider policy:

• The PR adds a new provider at plugins/memory/graphnosis/.
• AGENTS.md:786-795 and CONTRIBUTING.md:70-84 state that the in-tree plugins/memory/ set is closed; new backends must ship as standalone plugins.
• A standalone Graphnosis plugin can use the existing $HERMES_HOME/plugins/ discovery path (plugins/memory/__init__.py:109-119) and the same MemoryProvider interface, setup flow, and provider CLI registration.

Please publish the provider as a standalone plugin repository or pip entry point and share it through the Nous Research Discord #plugins-skills-and-skins channel. This is a coupling and maintenance policy, not a judgment of the integration's quality.

---

Closed as not-planned per standing maintainer policy (in-tree-memory-provider). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.