Proofline agents plan the research, gather from your documents and approved sources, check each claim and write a cited report your team can review. Completed jobs show their cost alongside the sources used.
Already ready to explore? Create a workspace ↗Registration ID: DEMO100. Submitted by the applicant.
The application and certificate contain different registration IDs.
Proofline is a web app and API for business research. Agents gather from named sources and your files, write a structured report, and a fact-check reopens every cited source before you rely on it.
Playbooks, agents and your own questions all run on the same engine: it plans, researches in parallel, writes and verifies. The sources it read and what it cost stay attached to every result.
PROOFLINE / HOW IT FITS TOGETHER
From your first question to a reviewed result. Follow each step, inspect the companies and people it finds, and keep control of what happens next.
Reads your website for products, value propositions, proof points and named customers.
A research workspace for teams turning scattered information into clear decisions.
Eight stages, each a cited research step. Work emails are found through the providers you connect or proven from public pages, and every draft waits in an outbox for you.
j.novak@example.comValidProviders are tried in your order until one verifies the address. Without a provider match, Proofline uses an address printed on a public page, or one built from a pattern that two published addresses at the domain prove. Never a guess.
Every find is kept and checked first next time, so you don’t pay a provider twice for the same person.
Finds made with your provider key stay with your workspace. A person removed on request is never saved or looked up again.
Research agents cite what they find. Then a separate fact-check, run by Proofline rather than the model, reopens every cited source and looks for the quoted passage.
A report is stored as claims, not just prose. Each claim keeps the page or file it came from, the passage quoted and a confidence level.
Acme Logistics GmbH has been registered in Hamburg since 2011.1
“Eingetragen: Amtsgericht Hamburg … seit 14.03.2011”
See the source. Inspect the passage. Review the finding.A matching quote is evidence to inspect, not a guarantee that a claim is true.
Ask for candidate screening, KYC and KYB checks, sales research, or market intelligence. Leave the composer on Auto and Proofline selects a suitable agent—or researches the question directly. Review the findings and sources in one place.
Check Example Freight Ltd before we sign them as a customer.
Company identity and ownership findings, source references, and unresolved checks.
Start from a gallery agent, installed once per workspace.
When an agent has test cases, a new version must pass them before it goes live. Skipping the gate is a deliberate, labelled choice.
The agents do the research. Your team defines the method, reviews the evidence, and makes the decision.
Start from an agent that already knows the work, then shape it to your process. Each one cites what it finds.
Track research, review approvals, and return to any conversation. Find the work you need, then pick up where you left off.
A useful finding should survive the conversation. Keep records, preserve their evidence, and bring the right context into the next run.
Turn a table from an answer into a tracker. Choose the columns your team needs, then keep the records together for future research.
| Company | Review | Next step |
|---|---|---|
| Northstar Demo | Needs evidence | Request certificate |
| Willow Example | Ready to review | Open brief |
| Cedar Sample | Research pending | Check ownership |
Sales, diligence, recruiting and market research each get a playbook made of gallery agents, and the sources that work needs.
Your website, the regions you sell in and what you offer.
Eight analysis tools work over retrieved or extracted data. Enable the ones your agent needs, alongside its research tools and your company procedures.
Rank candidate records and disclose identifier conflicts.
Calculate shares through supplied ownership edges; expose gaps.
Compare extracted fields with their original references.
Calculate supported ratios with formulas and missing inputs.
Compare contradictions without counting repeated sources twice.
Compare supplied snapshots and identify changed lines.
Check gathered evidence against the requested requirements.
Keep currency totals separate and flag duplicate candidates.
Define the brief and relevant sections. Inspect citations, request revisions and export Markdown or JSON from the report workflow.
The details that turn an answer into something you can act on.
Open the quoted passage, the original page, and the verification result. Contested claims stay visible.
The company is registered in Hamburg and has been active since 2011.1
One commercial dispute requires review.2
Proposed actions come with the context behind them. Approve when you’re ready.
Work with your own sources, workspace permissions, and a complete research trail.
Visit the Trust Center ↗Reviewer corrections turn into test cases. Make the next run better than the last.
Inside the platform ↗Models make mistakes. The system around them is built so those mistakes are caught, bounded and visible.
A run keeps money back for the write-up, so a tight budget still ends in a usable report rather than a half-finished one.
Anything that changes the outside world, from an email to a CRM update, waits for a person to approve it, unless your policy lets that one action run on its own.
When an agent has test cases, a new version must pass them before it goes live. If quality drops, it stays a draft. Skipping the gate is a deliberate, labelled choice.
When two sources disagree, both stay on the record with their evidence. Memory gains history; it never loses it.
Every job has a research budget and a time limit, enforced on the server. Nothing runs away.
If the machine running a job goes quiet, the run is finished from what was saved, completed as partial, or closed with nothing charged, and the run says which.
No integration project before the first answer. Sources, skills and agents are already wired in.
Catalogue sizes at the last count; they grow with every release.
Bring records, research, and your context together. Keep the source attached to every finding.
Official records. Public research. Your own context. Explore the sources behind the work, with references you can inspect.
Source coverage differs by country and record type. Email providers work only with your own account and key.
The hard parts of putting agents into production come built in.
Capped at $6.00 before it started. You pay what was used, never more.
Every run shows its steps, the sources it read and what it cost, against a cap set before it started. Your team controls access, review and delivery, and the audit log records who did what.
Plans, parallel researchers, fact-checks and hand-offs between agents, run for you.
Public catalogues, your files, connectors and MCP servers, in one place.
API keys limited to the scopes they need, passkeys and MFA for people, and SSO with SCIM on plans that include them.
Every job shows what it cost and which sources it used, line by line.
A tamper-evident log of who did what and when, ready to export.
Choose how long runs, files and reports are kept, and delete on your schedule.
What is built into every workspace today. No badges we haven’t earned.
Passkeys, authenticator apps, SMS codes and backup codes. Passwords hashed with Argon2.
Admin, Builder, Reviewer and Viewer, enforced by the API, not just hidden in the UI.
Hash-chained per workspace and kept at least 400 days. Export on plans that include it.
Per-scope access, optional expiry and IP allowlists. Secrets stored encrypted.
Google Workspace or OpenID Connect sign-in and SCIM provisioning, on plans that include them.
Keep files 30 days to 7 years; audit log 400 days to 7 years.
Not certified yet. Proofline does not hold SOC 2 or ISO 27001 certification. Our AI providers are named on the subprocessors page.
Read the Trust Center ↗Read it, send it, present it or feed it to your systems. Same evidence underneath every format.
A cited report in Markdown, with charts and tables
CSV or Excel from batches and trackers
Structured answers for your own systems
A share link that opens the report in a browser
Charts and tables over a batch or a tracker, with a share link
Rows written straight into a connected database
A spoken briefing of the findings
Bring research into your own workflow. Explore runs, reports, and structured evidence.
// Server-side: configure your API origin, live key and workspace.
// Persist the idempotency key with this request; reuse it for retries.
const origin = process.env.PROOFLINE_API_ORIGIN;
const workspace = "your-workspace";
const headers = {
Authorization: `Bearer ${process.env.PROOFLINE_KEY}`,
"Content-Type": "application/json",
"Idempotency-Key": crypto.randomUUID(),
};
const response = await fetch(
`${origin}/v1/workspaces/${workspace}/runs`,
{
method: "POST",
headers,
body: JSON.stringify({
question: "Research Acme Logistics and its supplier risks",
effort: "standard",
require_plan_approval: false,
}),
}
);
if (!response.ok) throw new Error(await response.text());
const run = await response.json();
console.log(run.id, run.status, run.charged_micro_usd);Give your coding agent this block. It reads the docs written for agents, connects the MCP server and installs the plugin.
Set up Proofline for this project. 1. Read https://proofline.iamrraj.com/llms.txt and https://proofline.iamrraj.com/.well-known/agent-onboarding.md. 2. Add the Proofline MCP server to your MCP config (.mcp.json): { "mcpServers": { "proofline": { "type": "http", "url": "https://proofline.iamrraj.com/v1/mcp", "headers": { "Authorization": "Bearer ${PROOFLINE_API_KEY}" } } } } 3. Install the Claude Code plugin from a checkout of the Proofline SDK: claude plugin marketplace add ./sdk/claude-plugin claude plugin install proofline@proofline Use an API key from Settings → API keys as PROOFLINE_API_KEY. Never commit it.
Bring the task, the documents and the outcome your team needs. Join the company pilot waitlist.
A focused pilot, with scope and acceptance criteria agreed together.
