Frequently asked questions
Short answers on the record, governance, residency and what runs today.
Is OIDA an assistant or a chatbot?
No. It is a memory that your existing AI clients query. You keep working in Claude Code, Cursor, Codex or the client you already use.
Does it replace our wiki, ADRs or document store?
No. Your original documents remain in their systems. OIDA stores the source text you explicitly submit, decisions, evidence and references; MCP answers return compact excerpts.
How is this different from RAG, agent memory or a knowledge graph?
Those systems store and retrieve text or facts. OIDA governs the state of decisions: which one is in force, what it replaced, where it applies, what contradicts it and when the record cannot say. It resolves that state deterministically, without a language model or similarity ranking, and it can sit above a retrieval layer rather than replace it.
What does ‘current’ mean here?
That the decision satisfies OIDA’s conditions on evidence, authority, confidence and absence of unresolved conflict. The MCP response calls this status verified. It is an application-level status, neither a digital signature nor a guarantee that the decision is correct.
Who approves a decision?
In reviewed mode, admins and owners approve decisions; in peer mode every member can. Ontology approval always needs an admin or owner. Decisions inferred from prose remain proposals until reviewed.
Where is our data?
In Frankfurt, in the database we run, with workspace separation enforced at row level. Clients authenticate with OAuth 2.1 and PKCE. OIDA issues no long-lived tokens. Audited writes create audit events; successful resolver calls create usage events. Optional capture records successful calls and results. For enterprise integration the registry stays in your own infrastructure, under your own policy.
Does OIDA train models on our data?
Not today. OIDA Cloud stores decisions, evidence and history and trains nothing. The OIDA Intelligence pipeline for datasets, evals and LoRA adapters is implemented and tested in a separate codebase and is not connected to OIDA Cloud. If your organisation chose to use it, training would start only from a dataset an owner had admitted under recorded rules, and a candidate would be compared with a baseline before an owner decided on promotion.
Can we run it ourselves?
Yes. OIDA Local is one TypeScript process, one SQLite file and one MCP server over STDIO, for a single operator on their own machine.
Which clients work today?
Claude Code, Cursor and Codex have completed a real run end to end. ChatGPT developer mode, Claude apps and GitHub Copilot in VS Code follow the vendor’s published path. Any client with Streamable HTTP and OAuth 2.1 can connect. Clients that only send a static header cannot.
Is OIDA Cloud ready for production use?
OIDA Cloud is live and in its pilot phase. OIDA Local is shipped. Talk to our team before routing production decisions through Cloud, so that sources, scopes and owners are agreed first.