Memory Page — 5-Tab Architecture

Each tab's API surface and on-disk data source; active tab persists in localStorage

MemoryPage /memory route · TanStack Query · active tab in localStorage Agent Memory Wiki Map Chat Settings Agent Memory tab per-agent markdown files agent selector: Hermes Switch Neo · Trinity · Morpheus file list + preview pane inline edit + save GET /api/memory/agents/:id/files PUT …/:id/files/:filename DELETE …/:id/files/:filename routed via Hermes gateway :8642 ~/.hermes/profiles/ <agent-id>/memory/*.md one markdown file per memory entry scoped per agent identity Wiki tab shared knowledge base list of markdown pages create / edit / delete pages readable by all agents GET /api/knowledge/pages PUT /api/knowledge/pages/:id DELETE /api/knowledge/pages/:id routed via Hermes gateway :8642 ~/.hermes/ MEMORY.md · memory/ · memories/ rooted at MEMORY.md index all agents can read Map tab D3 force-directed Memory Map nodes = gist · fact · wiki entries edges = ctx + reference links hover → title + connection count legend isolates node categories GET /api/memory/graph mnemosyne knowledge graph gist / fact / wiki node types routed via Hermes gateway :8642 mnemosyne graph (gateway) gist / fact / wiki + edges served as JSON by gateway force-directed layout in browser Chat tab RAG over wiki corpus type question → top-K retrieval streamed LLM response wiki chunks as LLM context POST /api/knowledge/search (top-K) POST /api/send-stream (RAG SSE) SSE tokens streamed to chat UI routed via Hermes gateway :8642 wiki corpus (vector index) ~/.hermes/ markdown as source top-K chunks injected into context provider: gateway session provider Settings tab wiki source + cache controls local filesystem vs GitHub wiki read + edit wiki source config clear knowledge cache per-agent memory → Profile wizard GET/PUT /api/knowledge/config POST /api/knowledge/sync?action=clear routed via Hermes gateway :8642 knowledge config + cache wiki source: local | github ~/.hermes/knowledge-config.yaml knowledge cache cleared via sync On-disk roots (all under ~/.hermes/) profiles/ <agent-id>/memory/*.md Hermes Switch · Neo · Trinity · Morpheus wiki root MEMORY.md · memory/ · memories/ all markdown · readable by all agents memory graph mnemosyne: gist · fact · wiki served via /api/memory/graph vector index wiki chunks for RAG top-K retrieval for Chat tab config knowledge-config.yaml source + cache settings Hermes gateway :8642 all /api/knowledge/* and /api/memory/* route here Hermes Switch UI · Memory Page · 5-tab architecture · /memory

Agent Memory tab

  • · Per-agent markdown files in profiles/<id>/memory/
  • · Agent selector: Hermes Switch, Neo, Trinity, Morpheus
  • · Inline edit + save via PUT /api/memory/agents/:id/files/:f
  • · DELETE removes individual memory files

Wiki tab

  • · Shared knowledge base readable by all agents
  • · Rooted at ~/.hermes/MEMORY.md index
  • · memory/ and memories/ subdirectories
  • · CRUD via /api/knowledge/pages/:id

Map tab

  • · D3 force-directed Memory Map from /api/memory/graph
  • · Nodes: gist, fact, and wiki entries
  • · Edges: ctx and reference links
  • · Legend isolates or mutes node categories
  • · Force-directed layout rendered in browser

Chat + Settings tabs

  • · Chat: RAG — top-K retrieval + /api/send-stream SSE
  • · Wiki chunks injected as LLM context
  • · Settings: local vs GitHub wiki source toggle
  • · Clear knowledge cache controls
  • · Per-agent memory config lives in Profile wizard
Hermes Switch UI · Memory Page · 5-Tab Architecture · all data under ~/.hermes/