About Folden
AI-native artifacts for regulated work.
Folden is an artifact platform for solo and small-firm professionals whose work is documents — NIL representation, cross-border tax, operations consulting. Drop one finished example. Folden learns your shape, your voice, your judgment from it. Every draft after that is produced as a complete artifact, ready for the small refinements that encode your specific judgment.
The shift
Generic AI is conversation-shaped. Your work isn’t.
Experts don’t want a chat partner. They want the artifact in front of them, mostly correct, ready to edit, send, sign. Folden reads your inputs the way you read them and hands you a draft you can refine — not a conversation you have to manage. Every edit is a lesson; every next case is closer to your voice.
How it works
How to transfer your expertise to AI without losing yourself.
Most expert judgment is tacit. You can’t write it down without losing it — that’s why prompt engineering breaks for anyone whose output isn’t a search query. Folden takes a different approach: instead of asking you to articulate your expertise, it watches you demonstrate it.
-
01
Show one finished artifact.
A contract you’ve signed off on, a memo you’ve filed, a runbook you’ve handed off. Folden reads it the way you read it — structure, voice, judgment, the patterns you’ve baked in.
-
02
Drop the next case’s inputs.
Folden produces the draft you’d normally write. Not a conversation; the artifact, ready to edit. You stay in the flow you already know.
-
03
Edit, and Folden learns.
Every edit is captured as a candidate pattern, scoped by horizon: long-term (always), conditional (this jurisdiction, this case kind), or one-off. You confirm the scope; Folden remembers. After a few cases, your private patterns library is a structured representation of the expertise that used to live only in your head.
-
04
Automate the structural work, keep the judgment.
What gets automated is the structural drudgery — formatting, boilerplate clauses, jurisdictional fit, audience tailoring. What stays yours is the judgment that makes you the expert. Folden is the apprentice that watches; you remain the protagonist.
If you’ve been wondering how to take what’s in your head and put it into AI — the answer isn’t to write it down. It’s to demonstrate it once and let Folden capture everything that comes after.
Built on Latten
Your client data was never a training dataset.
Folden runs on Latten, a sovereignty layer purpose-built for regulated work. Every case is encrypted in your space with a per-user key generated by AWS KMS. Every read and write is recorded in a cryptographic audit chain that you can verify yourself. PII is tokenized at the model boundary, so Claude, OpenAI, or any other provider never sees raw client names, identifiers, or financials — only structural placeholders. Right-to-erasure is a single-row delete: the encrypted key disappears, every ciphertext becomes permanently unrecoverable.
- Per-case encrypted vault
- Cryptographic audit log
- PII tokenized at the model boundary
- Scoped delegation tokens
- Instant revocation
Who Folden is for
Solo and small-firm experts. Today.
-
NIL & sports law
Athlete agreements, exclusivity, royalty floors, morals clauses, termination notice. Folden reads the offer, applies your standard clauses, and flags the terms you usually negotiate harder.
-
International tax
FY memos, transfer pricing, Pillar Two analysis. Folden reads the financials, applies your jurisdictional patterns, and drafts a memo that sounds like you wrote it — because the structure did.
-
Operations consulting
Runbooks for ops managers, vendor matrices for procurement, training decks for floor staff, status notes for the C-suite. One engagement, many audiences — same expertise, tailored output.
-
Coming soon
Litigation, M&A, healthcare compliance, more. If you produce a recurring artifact from variable inputs, Folden is for you.
Frequently asked
Questions experts ask before they trust a tool.
-
How can I transfer my expertise to AI?
Show it one finished example. Drop a contract, memo, brief, or runbook you've already produced — Folden reads how you structure it, the language you use, the patterns you apply, and the judgments you make. From that one example forward, every draft Folden produces sounds like you wrote it. Every edit you make becomes a captured pattern, scoped to whether it's a one-off preference, a conditional rule (this jurisdiction, this case kind), or a permanent rule (always). Your expertise transfers by demonstration, not explanation.
-
How do I automate my work without losing judgment?
By keeping the expert as the protagonist. Folden doesn't replace your judgment — it captures and amplifies it. You produce a finished example once; Folden learns your shape from it. You edit a draft; that edit becomes a pattern you can reuse. Over time, the first draft of every artifact is closer to your final version, but you stay in control of every decision. The work that's automated is the structural drudgery — the work that's preserved is the judgment that makes you the expert.
-
How do I take what's in my head and put it into AI?
Don't try to articulate it — demonstrate it. Most expert judgment is tacit; you can't write it down without losing it. Folden lets you show the AI what you produce, then watches how you refine it. Every edit you make is a piece of your judgment becoming explicit. After a few cases, your private patterns library — long-term rules, conditional ones, one-off preferences — is a structured representation of your expertise that the AI can apply on your behalf, in your voice, on every future case.
-
What is Folden?
Folden is an AI-native artifact platform for solo and small-firm professionals in regulated work. Experts drop one finished example artifact (a contract, memo, brief, runbook), Folden learns their voice and judgment from it, and every subsequent draft is produced as a complete artifact ready to edit and send.
-
Who is Folden for?
NIL representation attorneys, cross-border tax CPAs, and operations consultants helping foreign companies set up US or EU operations. More broadly, any solo or small-firm professional whose output is documents and whose judgment is hard to articulate but consistent across cases.
-
How is Folden different from ChatGPT or Claude?
Folden is artifact-shaped, not chat-shaped. Generic AI is conversation-shaped; expert work isn't. Experts want the artifact in front of them, mostly correct, ready to refine — not a chat partner that needs re-prompting from scratch every case. Folden also keeps client data sovereign: per-user encryption, cryptographic audit chain, and PII tokenization at the model boundary so Claude or any other LLM never sees raw client names, financials, or identifiers.
-
Can I trust Folden with regulated client data?
Yes. Every byte of client data is encrypted at rest with AWS KMS — neither Folden nor its hosting provider can read it without the user's active session. Every read or write is recorded in a cryptographic audit chain that the user can verify. The only place data travels is to the LLM provider, and even there PII is tokenized: Anthropic, OpenAI, or any other provider sees structural placeholders, never raw client names or numbers.
-
Does Folden train on my data?
No. The base model is never fine-tuned. Patterns captured from your edits live in your own encrypted vault and travel as prompt-time exemplars at draft time. Anthropic's API does not train on API traffic by default. Your data is yours; the model is shared.
-
What happens if I want to delete my account?
Right-to-erasure is a one-row delete. The encrypted DEK is the only key to your data. Delete the row and all ciphertext — including R2 backups — becomes permanently unrecoverable.
-
What is Latten?
Latten is the sovereignty substrate Folden is built on: per-user envelope encryption, cryptographic audit chain, PII tokenization at the LLM boundary, and scoped delegation tokens. Latten is what makes AI deployable in regulated work where chat-style tools cannot legally go.
-
Why artifact-shaped instead of chat-shaped?
Because expert work isn't conversation. A tax CPA producing a Pillar Two memo doesn't want to chat about it; they want the memo, mostly correct, ready for the small refinements that encode their specific judgment. Every refinement becomes labeled training data — Folden detects repeated edits across cases and proposes them back as captured patterns.