Review agents
Give your users built-in review agents, or let them create custom ones with plain English instructions. Findings land as comments a human accepts or rejects.
Stop launching AI features your enterprise buyers won’t approve.
Free tier. No credit card. First comment in 5 minutes.
Brand Agent
advisory · auto-runs on publish
Launch Your Review Agents — ship marketing your brand team already trusts. Every headline on-spec, every pricing claim cited. Plans start at $49/seat, no annual lock-in.
findings post as comments anchored to the exact text · approve to apply the fix
Headline uses title case. Brand guidelines §3.1 require sentence case.
Suggested fixLaunch Your Review Agents → Launch your review agents
Review agents running inside products at
What it is
Velt Review Agents are AI reviewers that check the content and data in your product before a human looks. Offer built-in agents to your users, or let them create custom ones with plain English instructions: review for brand violations, missing clauses, broken claims, policy conflicts. Seven built-in agents come with Velt, each individually toggleable: Spell Check, Grammar Check, Broken Links, PII Detection, Profanity Filter, Sensitive Data, and Consistency Check.
Headline uses title case. Brand guidelines (section 3.1) require sentence case.
cites brand-guidelines · §3.1Suggested fixLaunch Your Review Agents → Launch your review agents
“Good catch, also watch the subject line.”
resolved
Creator accepted the fix — content updated, agent reran and verified.
consent visible·every finding
How it works
Install the SDK, wrap your app, and enable agents for the documents you want checked. Findings land as comments in your product, with redlines and suggested fixes where relevant.
npm install @veltdev/react<VeltProvider
apiKey={VELT_API_KEY}>
<YourApp />
</VeltProvider>// enable agents for the document in the Velt Console first
<VeltReviewAgents
documentId="marketing-email"
/>An agent runs automatically on content submission, as a node in an approval workflow, or through the execution API. It reviews the target and leaves findings as contextual comments, with redlines and suggested fixes where relevant. Each finding carries Approve and Reject buttons: a human accepts the finding or rejects it. On accept, the finding becomes a durable comment or fires a custom action through your webhook. Prior runs are stored and inspectable, and with Memory enabled, agents read past decisions and stop re-flagging what the org has already settled.
// standalone on submission · workflow node · execution API
Teams that build it budget a quarter for the first version and keep paying for evals and model churn. The 3 steps above replace the quarter; the capability wall below replaces the long tail.
Skip the steps. Have your agent set it up.
npx -y @velt-js/mcp-installer
claude mcp add velt-installer -- npx -y @velt-js/mcp-installer
npx -y @velt-js/mcp-installer
npx -y @velt-js/mcp-installer
npx -y @velt-js/mcp-installer
Where findings anchor
APIs and pipes
Launch review agents this week.
No credit card. Works with React, Next.js, Vue, Angular, and HTML.
Capabilities
Each card is the live SDK. Toggle to Code for the exact snippet that renders it.
React
Next.js
Angular
Vue
HTML// seven built-ins, each toggleable
velt.agents.enable(["pii-detection"]);Spell Check, Grammar Check, Broken Links, PII Detection, Profanity Filter, Sensitive Data, and Consistency Check, each toggleable on its own. A legal product keeps PII Detection and drops the spell check that trips on jargon.
“flag any competitor mention”
velt.agents.create({
instruction: "flag any competitor mention",
});Type instructions like “flag any competitor mention”; the setup assistant sharpens them and proves them on test samples. Every team’s standards become agents, no engineering ticket.
Pricing claim on line 4 has no cited source: anchored here, threaded, AI-badged.
// posts through the comments API
<VeltComments />The agent posts through the comments API humans use: anchored to the clause or cell, threaded, AI-badged, with Approve and Reject on every finding. In the work, never a separate report.
velt.agents.update("pii", { mode: "blocking" });A toggle per agent: blocking findings gate content until resolved; advisory findings warn and record any override. PII removal is a gate; brand voice is a suggestion.
Headline uses title case. Brand guidelines require sentence case.
cites §3.1velt.agents.update("brand", {
knowledge: ["brand-guidelines.pdf"],
});Point an agent at the brand guidelines PDF or style manual stored in Memory; findings cite the exact section. The reviewer sees why, not just what.
Headline should be sentence case per §3.1. Proposed redline below.
// one tap accepts the redline
onFindingAccept(() => agent.rerun());Where the agent can fix what it flagged, it proposes the redline: one tap accepts, the content updates, the agent reruns to verify. Flag to fixed without a resubmit cycle.
// resubmit: fixed resolve, open persist
velt.agents.run({ match: "merge" });On resubmit, fixed findings resolve and open ones persist with their reply threads; only genuinely new issues notify. The second pass on a filing reads like progress, not a fresh pile.
// agent as an approval node
{ type: "agent", onFail: "route:specialist" }Any agent runs as a step in an approval pipeline: pass advances the deck or budget; fail routes to a specialist. First-pass review happens before a human’s queue fills.
velt.agents.fromChecklist("qa-300.csv");Upload a 300-item QA checklist; Memory normalizes it and the converter proposes focused agents, deduplicated, scoped to the right pages. Findings cite the checklist item they enforce.
Headline uses title case. Brand guidelines require sentence case.
{ "confidence": 0.91,
"checked": ["text"], "notChecked": ["images"] }Every finding carries confidence, the instruction or knowledge section behind it, and a declaration of what was and was not checked. Reviewers trust agents they can interrogate.
Agents pre-check every asset before a human opens it, and review time on first-pass items dropped by more than half. Reviewers spend their attention on judgment, not typos.
Review operations lead · content platform
Little big details
Shipped functionality only. The confirmed set leads; draft items render after engineering sign-off. This is the part of an in-house build that never ends.
Make it yours
Themeable AI badge and confidence display, white-label naming, and a full API surface: definition CRUD, execution, hooks, custom accept actions, and events.
AI badge distinguishable from human comments but fully themeable; confidence display configurable (percent, bar, or hidden); white-label, the badge shows the agent’s name, never “Velt AI”; findings inherit the comment dialog primitives for fully custom UIs; dark mode via the same variables.
Headline uses title case. Brand guidelines §3.1 require sentence case.
<VeltReviewAgentsWireframe>
// your finding markup, Velt reviews
</VeltReviewAgentsWireframe>Agent definition CRUD APIs, an execution API, the useAgentResults / useAgentConfig / useAgentAnalytics hooks, custom actions on accept through your webhook, and events on every finding.
velt.agents.create({ instruction });
onFindingAccept((f) => runAction(f));We white-labeled the agents under our own brand, and our users never see “Velt AI.” The consent step on every finding is what made our customers comfortable shipping it.
Product lead · AI-native SaaS
In production
Tabbed by vertical, with verified customer screenshots.
This Digital Sales Room gives you everything you need to move forward — from proposal to plan.
@Fin We need to make a working demo for Rene and his team
Brand, legal, and compliance agents pre-check every email and deck against the uploaded guidelines. The client sees work that already passed the standards, and reviewers open queues that are mostly green.For sales enablement
finance / receivables
Can we confirm the PO before approving this one? @Sarah
1 ReplyPII Detection and Sensitive Data run as blocking gates on filings and forecasts before any human review. The numbers reach the approver clean, with every agent check on the record.For fintech and FP&A
Cotton Supplier Ltd.
AgriCotton Farms
Village Kheda, Gujarat, India
SpinCo Yarns
18 Narol Rd, Gujarat, India
DyeWorks International
Calle de Mayo, Mexico
Can you attach the GOTS cert for this lot before we approve? @Sean
1 ReplyConsistency agents compare the phone number, hours, and address across every page and record, citing where they differ. The counterparty never catches the mismatch your own site missed.For operations
Privacy Risk Review
Wren discovered a feature request in Jira involving users' health data and triggered a review after analyzing the PRD in Confluence against company policy and US regulation.
JStarted by Juan MendezThis feature processes PHI, so HIPAA applies. I recommend a PIA and a vendor BAA review before any data flows. @Juan
Compliance agents pre-check every filing against the control library before a human opens it, flagging gaps as comments with Approve and Reject attached. Officers review queues that are mostly green, every check on the record.For compliance
Superflow
1 · Subscription & Fees
Customer subscribes to the Service on a monthly basis and shall pay the then-current fees in advance of each billing period. Fees are non-refundable except as expressly set out in this Agreement.
2 · Term & Auto-Renewal
The subscription renews automatically for successive one-month terms unless either party gives at least thirty (30) days' written notice of non-renewal before the end of the then-current term.
3 · Limitation of Liability
Except for breaches of confidentiality, each party's total aggregate liability under this Agreement shall not exceed the fees paid by Customer in the one (1) month preceding the event giving rise to the claim.
4 · Governing Law
This Agreement is governed by the laws of the State of Delaware, without regard to its conflict-of-laws provisions, and the parties consent to the exclusive jurisdiction of its courts.
@Jordan Auto-renewal needs 60 days' notice to match our policy, not 30.
@Maya Liability cap is one month of fees — Legal wants 12 months before we sign.
@Maya Governing law should be New York to match the MSA.
Clause agents screen each contract against your playbook the moment it lands, citing the exact term that drifts. Counsel reviews the exceptions, not the whole document, with a consent step on every finding.For legal
Make the hero headline punchier and add a primary CTA.
Done — rewrote the headline, tightened the subtext, and added a primary CTA in hero.tsx.
Add a logo strip with social proof under the CTA.
Added a five-logo trust strip below the CTA and wired it into hero.tsx.
Ship your app in a weekend
Get startedUse our brand orange on the CTA and cut the headline to one line. @Sean
Agents review generated drafts the moment they land, flagging issues as comments with Approve and Reject attached. Your users trust the AI because a consent step sits on every finding.For AI-native SaaS
See it running in products like yours.
30 minutes, with an engineer, not a sales deck.
Related primitives
Comments carry the findings, Memory grounds them, and Approval flows route them.
Findings are comments, the same threads your users already know.
Findings are comments: the same threads your users know.
3 RepliesKnowledge sources and the precedent agents read before flagging.
knowledge sources · the precedent agents read before flagging
Agents run as workflow nodes with pass/fail routing.
Simple built-in agents (spell, grammar, PII) run inside Velt’s GCP; the consistency built-in and custom agents use a frontier model under a zero-data-retention DPA, with a fully GCP-resident routing option per customer. Agent findings follow the self-host data provider model, so finding content can stay on your infrastructure. See self-hosting and governance.
Book DemoProof
A consent step on every finding is what cleared the security review.
0 blockers
Our enterprise buyer’s security team approved the AI feature on the first review. Every finding had a consent step and a judgment record behind it.
2x faster
Brand and legal queues open mostly green now. The agents pre-check every deck against our guidelines before a human ever sees it.
10s
Seven agents run in parallel in under ten seconds, and only genuinely new issues notify on a resubmit. Reviewers finally trust the second pass.
FAQ
Seven, each individually toggleable: Spell Check, Grammar Check, and Broken Links for quality; PII Detection, Profanity Filter, and Sensitive Data for sensitivity; Consistency Check, which uses Memory and stays off until Memory has data. Built-ins use the same framework as custom agents, so disabling one never affects another.
Install @veltdev/react, wrap your app in VeltProvider with your API key, and enable agents for the documents you want checked; configure which agents run from the dashboard. Or have your coding agent set it up through the Velt MCP installer.
Yes. A PM writes plain English instructions, the setup assistant rewrites them into precise checks and proves them on test samples, and the PM marks each right or wrong until the agent behaves. No code, no prompt engineering.
Everything is isolated per tenant: agent definitions, findings, and knowledge are scoped to your API key, with physically separated vector store namespaces, and your content is never used to train models. Custom agents call a frontier model under a zero-data-retention agreement; simple built-ins (spell, grammar, PII) stay on Velt’s GCP; the consistency built-in routes like custom agents, and a fully GCP-resident routing option exists for stricter requirements.
Only if you make it blocking. Agents default to advisory: findings show as warnings, reviewers can proceed, and overrides are recorded. Blocking agents gate content until findings resolve, and an authorized override per item is always available.
A human rejects the finding and the override goes on the record, which feeds agent tuning. Findings below 50 percent confidence never render, citations are verified before display, and identical reruns return identical results. If infrastructure fails, content proceeds; reviews never block on Velt.
Yes. Any agent runs as a node in a Velt approval workflow: pass advances the work; fail routes it to the right human. The same agent can also run standalone on every submission, with no workflow at all.
Yes. Upload guidelines to Memory and point agents at them; findings cite the exact section. A QA checklist converts into focused agents that cite the checklist item they enforce.
Velt is priced on usage, not seats: you pay for documents with review activity in a month, with a free tier for development and early production. How built-in and custom agents are packaged within that is being finalized.
Free tier. No credit card. First comment in 5 minutes.
30 minutes, with an engineer, not a sales deck.
Headline uses title case. Brand guidelines §3.1 require sentence case.
Suggested fix
Launch Your Review Agents→ Launch your review agents