hypernexus

HyperNexus Integration

HyperNexus is a local AI control plane running a Go kernel on port 7778 (dashboard 7779) with persistent multi-tier memory (L1 scratchpad, L2 vector store, L3 cold archive), semantic tool discovery across 60+ MCP servers / 26,000+ tools, knowledge graph, session import, and agent orchestration.

When to Use

• Start of any multi-step task → harvest prior context (hypernexus_context_harvest / hypernexus_search_memory)
• After a decision, gotcha, or non-obvious finding → store it (hypernexus_add_fact / hypernexus_record_observation)
• When you need a capability you don't have → discover tools (hypernexus_mcp_search / hypernexus_tool_discover) then route (hypernexus_mcp_call)
• When the user asks about system health, billing, or the repo → hypernexus_system_status, hypernexus_billing_status, hypernexus_repomap

MiMo Desktop Tools (native)

Tool	Purpose
hypernexus_search_memory	Keyword search over persistent memory
hypernexus_memory_vector_search	Semantic / conceptual memory recall
hypernexus_memory_list	List recent facts
hypernexus_add_fact	Store a fact / decision
hypernexus_record_observation	Record a knowledge-graph observation
hypernexus_knowledge_graph	Add entities/relations or search the graph
hypernexus_context_harvest	Pull relevant L2 context for a task
hypernexus_session_search	Search imported sessions
hypernexus_session_context	Get/set session goal & objective
hypernexus_mcp_search	Semantic MCP tool search
hypernexus_tool_discover	Broad capability discovery
hypernexus_mcp_call	Route a call to a downstream MCP tool
hypernexus_code_search	Pattern / AST / semantic code search
hypernexus_repomap	Repository map with symbols
hypernexus_system_status	Kernel + memory + MCP health
hypernexus_billing_status	Provider quotas & costs

HTTP API (port 7778)

# Memory
curl -X POST http://localhost:7778/api/memory/facts/add -H 'Content-Type: application/json' \
  -d '{"title":"...","content":"...","tags":["..."],"namespace":"project"}'
curl -s "http://localhost:7778/api/memory/search?query=<term>&limit=<n>"

# MCP discovery & routing
curl -s "http://localhost:7778/api/mcp/tools/search?query=<natural+language>"
curl -X POST http://localhost:7778/api/mcp/tools/call -H 'Content-Type: application/json' \
  -d '{"serverName":"...","toolName":"...","arguments":{}}'

# System
curl -s http://localhost:7778/health
curl -s http://localhost:7778/api/system/status
curl -s http://localhost:7778/api/billing/status

MiMoCode CLI Commands

• /hn-store — store a memory fact
• /hn-search — search persistent memory
• /hn-status — system status
• /hn-tools — discover MCP tools
• /hn-session — session context / search

Best Practices

1. Recall before you write: search memory before large tasks.
2. Store immediately: non-obvious findings decay fast — persist them when found.
3. Discover then route: use hypernexus_mcp_search before hardcoding tool names.
4. Graph what matters: entities + relations compound into durable knowledge.
5. The Go kernel is the source of truth — prefer the HTTP API over shell improvisation.