OIDA is an ontology framework designed to train local, proprietary, governed AI models on the signals your organisation produces every day. It turns unstructured company knowledge — decisions, evidence, hypotheses, tacit expertise — into a computable, queryable graph of typed epistemic objects. Built by Kakashi Ventures.
Every organisation runs on knowledge — decisions, evidence, hypotheses, dependencies, tacit expertise — but this knowledge is scattered across documents, chats, decks, tickets and people's heads. Teams cannot answer basic questions: Is this decision still active? Has anything contradicted it? How much should we trust this? Who knows about this? What blocks this?
Traditional tools — wikis, knowledge bases, search — treat knowledge as documents to find, not as structured objects to reason over. As a result, organisations keep rediscovering the same things, executing plans based on outdated assumptions, and losing expertise when people leave.
OIDA treats organisational knowledge as first-class computable objects. Every decision, piece of evidence, hypothesis, question or dependency is stored as a typed node — an epistemic object — with explicit relationships (supports, contradicts, implements, supersedes, blocks, derives from) and computed properties (importance, confidence, freshness, urgency, controversy).
The ontology defines nine epistemic classes (Decision, Constraint, Narrative, Evidence, Evaluation, Observation, Plan, Hypothesis, Question), each with its own commitment strength and temporal decay behaviour. A deterministic Knowledge Gravity Engine recomputes importance every six hours from real usage, evidence, contradictions, and time — no manual curation.
The signals the framework produces — classifications, corrections, retrieval patterns, contradictions, decay curves — are the training substrate for a local, proprietary, governed AI model. Hardware, training, and inference stay inside your infrastructure. Data, training signals, and the model itself never leave. The vendor cannot silently retune your context every quarter, because the context is yours.
OIDA ingests organisational signal from existing sources — docs, meeting notes, tickets, chats, CRM, code — and extracts epistemic objects with their relationships. The Knowledge Gravity Engine maintains them over time: applying decay, propagating contradictions, migrating objects between memory zones (Core, Working, Peripheral, Dormant). Hybrid retrieval then combines structural, semantic, and topological similarity, weighted by each object's current importance. Read the deep-dive →
It means turning the implicit knowledge that runs a company — decisions, rationale, evidence, dependencies, tacit expertise — into explicit, typed, queryable objects with relationships that machines can reason over. Not just search: structured inference.
No. Wikis store documents; OIDA stores epistemic objects. A document is a file you read; an epistemic object is a typed node with relationships, computed scores and temporal state.
LLMs are pattern-based; OIDA is ground-truth-based. The framework provides the organisational memory that AI agents need to reason correctly inside a specific company — and the training signals that a private, governed model can learn from over time.
OIDA is built by Kakashi Ventures. The technical foundations are described in the research paper.
Visit the contact page or write to hello@projectoida.com.