Control
Set the boundary before AI runs.
Folders define the files in scope. Execution domains define where work may run. Ordered rules choose from the models allowed inside that boundary.
Data classification · Guardrails · Routing
Flow for enterprise
Govern AI where the work happens.
Flow puts scope, routing, approval, evidence, and cost inside the document. Every approved run leaves a receipt your team can inspect.
8 adoption patterns are built and running today. Knowledge mining remains a clearly marked direction.
One governance system
Control. Verify. Learn.
The record is created as work moves from a bounded request to an approved artifact. Governance is part of the path, not a report assembled later.
Verify
Make every decision inspectable.
Exact diffs preserve human approval. Evidence states what was measured. Receipts bind the model, route, cost, checks, and saved bytes to one run.
Attribution · Evidence · Allocation
Learn
Keep the work as the record.
Plain files remain authoritative. Approved work can become structured knowledge without a hidden observation system or a second database to reconcile.
Curation · System of record · Knowledge mining
The proof pattern
One run. Four inspectable decisions.
-
01
Scope
Which files may be read.
-
02
Route
Where the model may run.
-
03
Approve
What a person accepted.
-
04
Receipt
What ran, changed, and cost.
Read the nine adoption patterns
1 · Allocation
TodayHow do we know which users need more AI and which need to cut back?
Every action carries its own cost record. Local runs record no charge because none was owed. Billed runs record the observed charge, never an estimate. Spend becomes a per-person, per-action fact you can read, not a monthly surprise.
2 · Data classification
TodayHow do we control which documents AI may touch, and which AI may touch them?
The folder is the classification. Flow only sees folders you open, and a document cannot reach outside its folder by naming a path. Model work runs on this Mac or through a cloud API provider you choose. A provider key turns its cloud route on; a runtime pointed at another machine is refused.
3 · Attribution
TodayHow do we know who changed which part of a document, human or AI, and when?
Daytime Agency proposes an exact diff for approval before saving. A standing Night Shift job applies the bounded changes you authorized in advance, marks them for Keep or Revert review, and records what ran. Version history keeps the changes readable later.
4 · Guardrails
TodayHow do we apply guardrails at the source of knowledge work?
Checks run inside the document, at creation, as named rules with six honest states. An override never turns Failed into Passed. The decision is recorded and the failure stays visible. Rules live where the work happens, not in a policy PDF.
5 · Evidence
TodayHow do we compare the quality of one AI generation against another?
Evidence is scored only where a measurement exists. A proofread reads "2 dimensions assessed, 3 need sources". A summary is deliberately not scored, because Flow refuses to print a number with no measurement behind it. When the same model writes and judges a change, the receipt says so.
6 · Routing
TodayHow do we route each part of the work to the best model, provider, and place?
The domain is the decision; the provider is a detail. Long documents are improved part by part, so each part gets a right-sized run. A fallback that would leave your Mac stops and asks. Routing reads measured capability, never a vendor claim.
7 · Curation
TodayHow do we curate structured value out of unstructured documents?
Convert Text to Table builds real structure from prose, and the review reports what was actually built, read from the proposed table itself, never from the model’s claim. The file stays plain Markdown, so the structure belongs to you.
8 · System of record
TodayHow do we keep one system of record feeding many documents?
Plain files are the only authority. Search results carry anchors that are re-checked byte for byte before Flow highlights a passage, so a citation cannot silently drift. The index is disposable; your documents are the record.
9 · Knowledge mining
DirectionHow do we capture how our best people decide, without "big brother watching"?
The receipt is made by the worker, artifact by artifact, as a side effect of work they already wanted to do. They consent by name and get the trust benefit of the same record. That is the only capture a professional volunteers for.