For fintech and FP&A
Numbers ship with names attached. Velt embeds cell-level comments, approval chains with quorum, immutable audit records, and review agents for mechanical checks in products selling into regulated finance.
Free tier. No credit card. First comment in 5 minutes.
finance / receivables
Can we confirm the PO before approving this one? @Sarah
1 ReplyThe review reality
If your product is a review, policy management, regulatory filing, or risk and attestation platform, that page is /for/compliance. This page is for products where the numbers are the work: budgets, forecasts, models, the close.
A regulated deal stalled on “who approved this number?”, and the answer was a reconstruction from old email threads.
The security questionnaire asked how your product proves sign-off on the board pack. The honest answer was screenshots.
The bank's vendor review stalled on “where does our data live?”, and your roadmap had no field-level answer.
An enterprise buyer asked for approval workflows on budgets, and the close calendar could not wait for the quarter it would take to build.
The loop
Velt runs a quarterly forecast through the full review loop inside your product: comment threads on the cell, a review agent's mechanical first pass, a staged approval chain with committee quorum, and an immutable record of every decision. Both humans and agents act through the same primitives, and nothing in the model changes without a human accepting it.
// one forecast, four steps, one immutable record. The committee signs off without leaving your product.
The analyst submits the forecast. The CFO's question about Q3 lands on the Q3 cell itself: threaded, attributed, resolvable. Threads anchor to spreadsheet cells and tables, so feedback never detaches from the number it is about.
Why is Q3 travel over plan? @Maya
A review agent runs the mechanical pass: variance thresholds, tie-outs between sheets, rates against contract, missing variance notes. Findings land as cell comments with Approve and Reject attached. Nothing in the model changes.
Travel 18% over plan; variance note missing.
The forecast enters the approval chain your product's users defined: the FP&A lead first, then a two-of-three finance committee quorum, then the CFO. Comment-gated steps hold the pipeline until every open cell thread resolves, and a reject branches it back with the note attached.
Every transition is timestamped and attributed: who was asked, who responded, what they decided. The auditor's sample request is a filtered API query that returns the complete chain, and the audit package assembles from the trail, not from old email threads.
Feature map
Each card links its feature page. Preview the UI, or read the code that renders it.
activities/getPOST /v2/activities/get
{ "data": { "organizationId": "your-org-id", "documentId": "q3-forecast" } }Every approval on a budget, forecast, or model, timestamped and attributed. The auditor's sample request is a query, not an investigation.
Explore Audit trail<VeltApprovalFlow
stages={["fp&a", "committee", "cfo"]}
/>Staged sign-off your users define in your UI on Velt's APIs: sequential steps, committee quorum, rejection paths. The close calendar runs as a pipeline, recorded.
Explore Approval flowsAnchored to the Q3 cell: not a note about it.
1 Reply<VeltComments />Threads on the cell, not about it. The question about Q3 sits on the Q3 cell: attributed, resolved, on the record.
Explore CommentsVelt.setDataProviders({ comment: commentDataProvider })Comment content on budgets and forecasts persists to your database; Velt keeps minimal identifiers. The vendor security review gets a field-level inventory, not a promise.
Explore Self-hostingQ3 travel is 18% over plan: variance over the declared threshold, note missing.
velt.addReviewAgent({
instructions: "flag variance over plan",
});The mechanical pass before a human looks: variance thresholds, tie-outs, rates against contract. Findings land as cell comments a reviewer accepts or rejects.
Explore Review agents<VeltMemory />Past decisions surface as precedent, so close number ten is reviewed like close number one, even as the finance team turns over.
Explore MemoryAgent action layer
In a financial product the stakes are the forecast, the model, and your customer's capital, so no agent holds write access to any of them. An agent finding arrives as a comment with the proposed action attached: “Vendor rate is 12% over contract. Suggest correcting line 7.” A named human approves or rejects. On approve, the change applies through your webhook with a permanent record of who allowed what; on reject, nothing happens and the rejection is logged. For higher-stakes actions, a journal adjustment or a close entry, escalate the consent: from one click, to committee quorum, to type-the-name confirmation.
Proposes · vendor rate is 12% over contract. Suggest correcting line 7.
Audit entry
Approved · CFO · Tue 09:14 · line 7 corrected · agent never held write access
on reject, nothing touches the model and the rejection is logged
In production
Customers in this space include FP&A, budgeting, and financial modeling platforms running cell-level review, staged sign-off, and audit records on budgets, forecasts, and models inside their own products. They embed comments, approval chains with quorum, and the audit query their regulated buyers ask for, instead of reconstructing who approved what from old email threads.
See it running in products like yours.
30 minutes, with an engineer, not a sales deck.
Compliance
Your buyers sell into regulated finance, so the strip leads with the certifications their vendor review asks for first.
The report is available under NDA, the evidence the bank's vendor review opens with before it looks at anything else.
GovernanceHIPAA support with a Business Associate Agreement for products that touch protected information alongside the numbers.
EU data residency options, and per-feature data providers that keep comment content on budgets and forecasts on your infrastructure while Velt keeps minimal identifiers.
Self-hostingArticle 14 of the EU AI Act requires demonstrable human oversight for high-risk AI systems under Annex III, credit and insurance among them, enforceable from August 2, 2026. If your product or your customers operate in that scope, Velt provides the mechanism, approval before action, and the evidence, a record of who approved what, when, and why. Talk to your counsel about whether you are in scope.
FAQ
Yes. Register the comments data provider and thread content on budgets, forecasts, and models persists to your database; Velt stores only minimal structural identifiers. The Complete Field Inventory documents every persisted field on both sides, ready for the vendor security review. See /self-hosting.
Comments, edits, and approval state changes are recorded automatically and queryable by document, user, or time range through the REST API, exportable as structured JSON. Approval records capture who was asked, who responded, and what they decided, with the note attached. See /audit-trail.
Yes. Threads anchor to spreadsheet cells and tables, charts, documents, and your own grid components, so the question about Q3 sits on the Q3 cell, attributed and resolved on the record. See /comments.
Yes. Workflows support sequential and parallel steps and quorum rules (where two of three must approve), with rejection paths and a full history per run. Approval flows is in beta. See /approval-flows.
Article 14 applies to high-risk AI systems defined in Annex III, credit and insurance among them, enforceable from August 2, 2026; it is not a universal obligation on every AI feature. If your product or your customers operate in that scope, Velt provides the oversight mechanism and the audit evidence. Talk to your counsel about whether you are in scope; see /governance.
Usage-based on monthly active documents: you pay for the workbooks, models, and board packs with review activity in a month, not per seat. There is a free tier for development and early production.
Free tier. No credit card. First comment in 5 minutes.
30 minutes, with an engineer, not a sales deck.