Decision Graph

Enterprise AI agents are intelligent but amnesiac. Every session starts from zero — no memory of past decisions, no institutional context, no understanding of what failed last quarter. Companies deploy agents that hallucinate because they have no grounding in company reality. DecisionGraph is the institutional memory OS that fixes this. It sits between your AI agents and your enterprise knowledge, giving every agent permanent, compounding memory grounded in your company's own documents, decisions, and discussions. Documents get ingested through a GraphRAG pipeline that extracts entity-relationship triples, builds a 3-layer hierarchical knowledge graph with community detection, and indexes communities for fast beam traversal. Every agent answer is stored as a permanent decision node with confidence scoring, outcome tracking, and causal links. Decisions age and decay when unused. The graph stays accurate, not just large. For major decisions, MiroFish Studio stress-tests them before commitment — simulating reactions across investors, competitors, and regulators — returning a consensus score stored as part of the decision record. Any agent — Claude, Gemini, GPT, LangChain — connects via MCP in five minutes. Nine integrations handle data flow including Google Drive, Gmail, and Supabase. Every company gets a completely isolated workspace proven by 13 cross-tenant isolation tests. Built on Gemini 3 Flash, NetworkX, sentence-transformers, and FastAPI with 40+ endpoints and durable background processing. Compliance queries drop from 3 days to 4 minutes. Legal research time cuts by 80%. Unlike standard RAG that stays static, DecisionGraph compounds — every agent interaction makes the next one better. Memory that compounds beats intelligence that forgets.