exploring-knowledge-graph

Exploring the Knowledge Graph

Explore Memoo namespaces through Memoo MCP tools (Memoo REST API only).

When to Use

Deep graph exploration:

• Tracing how a decision evolved over time
• Understanding connections between components
• Finding related architectural patterns
• Investigating why something was built a certain way

Namespace context

Prefer a session default (UUID or name — gateway resolves via GET /api/v1/namespaces/{id} detail at boot):

export MEMOO_NAMESPACE="0b00322d-f6ed-45b7-a2e8-b7059f71de34"
# or MCP args: --memo-namespace <uuid|name>

If a default is already configured, omit namespace on tool calls and skip memoo_list_namespaces unless the user asks to switch/list other namespaces.

Exploration Strategy

Step 1: Find starting point

memoo_search / search:
  query: "authentication architecture"
  limit: 5

Step 2: Explore relationships

Use entity UUIDs from search hits:

memoo_graph_traverse:
  entity_uuid: "<uuid from search>"
  max_hops: 2

Optionally ask for a synthesized view:

memoo_ask:
  query: "What depends on authentication and how?"
  use_reranker: true

Step 3: Get episode context

memoo_list_episodes:
  page: 1
  page_size: 20

fetch:
  id: "<episode_id>"

For a point-in-time slice:

memoo_temporal_query:
  query: "authentication"
  at_time: <unix_timestamp>
  limit: 10

Ingest (when asked to store knowledge)

1. Summarize first (prefer ≤300 words). Hard max is plan entitlement episode_content_words (API-enforced).
2. Call only memoo_create_episode via Memoo MCP — do not invent curl/REST bypasses.
3. Create is always async: expect job_id immediately (no episode_id on create). Poll memoo_get_job_status until completed / failed (ingest itself can take minutes).
4. On HTTP 413 / content too long → summarize further and retry.
5. Prefer the session default namespace (name). Do not pass raw UUIDs in tool args unless the host has no default.
6. If create fails, retry the MCP tool — do not shell out to the REST API or print API keys.