Querying the graph
You can query the KG when an agent needs durable context before writing or validating work. Start with specific read tools; use Cypher only when the dedicated tool does not cover the question.
Natural language queries
okto_pulse_kg_query_natural performs deterministic retrieval over embeddings, HNSW indexes, and traversal. It does not ask an LLM to invent an answer.
okto_pulse_kg_query_natural(board_id="...", query="why did we choose the current auth model?")
Cypher queries
okto_pulse_kg_query_cypher is read-only and intended for operators. Use explicit limits and narrow patterns.
MATCH (d:Decision) RETURN d.title, d.relevance_score LIMIT 20
Semantic search
HNSW indexes cover Decision, Constraint, Requirement, Entity, and Learning nodes. Semantic search is retrieval; it is separate from the stored relevance_score formula.
Decision history retrieval
Use okto_pulse_kg_get_decision_history when the topic is known but the decision ID is not.
Contradiction detection
Run okto_pulse_kg_find_contradictions before introducing a spec that may conflict with old decisions or constraints.
Global cross-board queries
okto_pulse_kg_query_global searches the global discovery graph and respects board access.
Relevance scoring
The v0.3.3 base score is:
clamp(0.4*source_confidence + 0.3*log(1+degree)/log(100) + 0.3*decayed_hits - contradict_penalty + priority_boost, 0.0, 1.5)
+0.3 to stored score and clamps at 1.5. It is an operator override, not a change to semantic retrieval.
Related pages
Last modified on May 8, 2026
