About Kodori
Agents aren't a feature. They're the substrate.
Every incumbent document system was built when "document management" meant storing documents. We think that premise is obsolete.
Every incumbent DMS — iManage, NetDocuments, SharePoint, FileHold, Documentum, OnBase, Procore — was designed when document management meant storing documents. Search was a feature. Classification was a form. Retention was an admin screen. Humans did the work; the software held the filing cabinet.
Generative AI makes that division obsolete. Once agents can read, classify, extract, reason over, and propose actions on documents at human-level accuracy, the bottleneck is no longer finding the file — it's having a document system agents can operate safely, transparently, and reversibly.
Kodori is that system. Every operation — create, classify, rename, move, retain, legal-hold, share, delete — is a typed tool with a Zod schema, an audit event, and a reversal path. The same catalog serves the internal agent loop and an external MCP server, so customers can integrate their own agents on equal footing.
- 01
Event-sourced from day one.
Every consequential mutation appends to a signed, hash-chained log. Current state is a projection. The compliance posture every regulated buyer asks for, built into the foundation rather than bolted on for an audit.
- 02
Metadata-first, collections-as-views.
No physical folder tree. Cabinets, drawers, matters, projects are all typed collections over the same metadata layer. One document can live in many contexts without duplication.
- 03
Reversible by construction.
Agent transactions can be previewed and reverted within the retention window. Destructive actions are tombstones, not deletes. An agent that drops the wrong bomb is recoverable.
- 04
Permission-trimmed at the index.
ACLs are applied at retrieval time, not in the UI layer. Deny always wins. Ethical walls hold at the database edge, not at a hopeful filter downstream.
Current focus is Phase 1: real capture, extraction, and workflow on the canonical architecture — with paying customers in both legal and AEC segments by month 24 of the project plan. Phase 4 ships the two vertical modules in parallel.
Longer horizon, each vertical reusing the same substrate: QMS in manufacturing, accounting firms, healthcare mid- market. Every vertical adds at most ~30% net-new code on top of the previous; common substrate first, modules second.
We build in public on Vercel. The code that ran this week is the code we committed this week. We prefer showing the product over talking about it.