Frequently asked questions

Sign in with Google, GitHub, or a 6-digit email code (no passwords). We create an isolated Sandbox workspace and catalog for you. No card required. Then paste a price book on Score or go straight to Onboard or your workspace. Enterprise SSO (SAML / OIDC) is scoped in the Enterprise order form, and is not required for Sandbox or Pro.

A real price book or packaging sheet: product names, SKUs if you have them, list prices, currencies, billing periods, and any seat or region rules. Messy text is fine. Atlas structures offers and flags conflicts before you publish.

Use the readiness score in the workspace. It highlights missing prices, conflicting offers, weak packaging, missing quote authority, and other blockers. Raise the score, then publish when the catalog is good enough for agents to trust.

Publishing versions your approved commercial truth and projects it onto the protocol rails your plan allows (see which protocols). Agents only see the version you publish.

Yes. Most sellers extract, score, and publish first, then connect agent traffic later. Sandbox is built for that dry run.

No. You keep your existing stack. On Pro, connect Stripe or Chargebee in Integrations, or any other system through the signed webhook; what happens after approval is below. Import Stripe or Chargebee products into a catalog draft (you still publish). Save a HubSpot private app token to write quotes as deals and access or deal-desk items as tasks, and to read six eligibility fields. Salesforce is next; join the waitlist in the workspace. Stripe connections accept only restricted (rk_) keys. Tax and fulfillment stay with your existing providers. Enterprise can honor signed contract pricing.

Atlas sends the approved quote to your billing system (Stripe, Chargebee or any billing system via signed webhook). Your processor charges the buyer, exactly as it does today. Above your threshold, a person on your team approves the quote first. Below it, the quote only goes to your billing when the buyer’s own human has authorized payment, and only if you gave that buyer’s key the quote:execute scope, which is off by default. Atlas refuses payment authorizations that are expired or capped below the quote total, and enforces each key’s hourly and daily handoff limits. If you register your issuers’ public keys (ES256, RS256 or EdDSA), Atlas also verifies each authorization’s signature, requires an expiry, amount cap and unique ID, and stops one authorization covering two quotes; you can require signed authorizations for every handoff. Your processor stays the authority on the payment. Sandbox uses Mock mode so you can rehearse the handoff without live charges.

Atlas holds your approved pricing, packaging, and policy in a versioned, evidenced catalog. Agents read that catalog on the rails that exist today. Adapters can change as protocols version. The catalog does not.

API-first and AI-native product teams, and developer-tool vendors who need catalogs agents can discover and price without scraping PDFs or guessing list prices. Marketplace operators, distributors, and broader quote-led B2B software are on the rail next.

Your commercial truth is structured, versioned, readiness-scored, and served on protocol rails agents can consume without one-off integrations, including a published quote mandate. Pro adds evidence-linked provenance and production rails. A catalog is agent-ready when its readiness score is 85 or higher.

A PairRail badge for catalogs that are published, score 85 or higher, and have no publication blockers. Pro and Enterprise catalogs earn it. A Sandbox catalog that passes the same checks gets the Sandbox badge instead. Agents get its live prices, but its quotes are tests that never reach live billing. If you pause the live catalog, the public badge switches to Paused until you publish again. Check any seller on Verify. It is not a third-party certification.

No. Quotes stay indicative until they are approved and handed off to your own billing. Atlas publishes your quote authority (that quotes are indicative and nothing binds before that handoff) on UCP and in quote envelopes.

In the workspace, set the quote cap agents can work within, the cap above which a manager approves, and your discount fences. Save them to the catalog draft and publish. Per-key limits apply on top, without republishing; see agent access.

Only the outcome. Each pricing answer or quote tells the agent whether it is indicative, needs approval or is denied, the reasons, who must approve and what it can do next. Your autonomous cap, manager cap and automatic discount limit are never in an agent response, on REST or any protocol rail.

No. Quotes an agent key got without approval count together toward your autonomous cap over a rolling window, 24 hours by default. Once the combined total crosses the cap, the next quote goes to your deal desk like one large order would. Only that key’s auto-approved quotes in the same currency count; quotes a person approved, and revoked or expired quotes, do not. Change the window in the workspace under Human-approval tiers, or set it to 0 to turn the check off.

Fail-closed exception handling. When a request exceeds the authority you published, Atlas returns a coded error (revise, poll for approval, or back off) and queues the ask for a human instead of guessing. Pro includes the deal-desk queue. Enterprise can require two or more different approvers on quotes above your manager max.

Plan Quota meters raw computational queries consumed against your monthly allowance (e.g. 25,000 queries on Pro across catalog discovery, price calculation, and readiness inspections). Atlas Insights puts that usage in commercial context: metered agent queries by type, recorded agent events by protocol rail when available, how many offers are gated from anonymous agents by disclosure level, agent searches that matched no public offer, and quotes waiting on or resolved by your deal desk. Insights only shows what was recorded. It never fills gaps with estimated or sample traffic.

No. PairRail is never the merchant of record and never handles money. Buyer agents get quotes and approval rules from us; payment happens through your existing billing.

Sandbox is free: one catalog live on MCP for discovery and indicative pricing, a Sandbox trust badge on the same checks, 3 Deep verification runs a month, quotes and billing handoff as a marked rehearsal with Mock mode, and a hard 1,000 queries/month cap. Pro unlocks production protocol rails, hands approved quotes to your own payment rails or a signed webhook, includes 25,000 queries/month, and keeps serving agents with soft overage instead of a hard cut.

Pro is $99/month, or $79/month when billed yearly ($948/year), billed in USD. Overage is $2 per 1,000 queries after the included volume. Choose monthly or yearly on the pricing page before checkout.

Yes. Choose monthly or yearly at checkout. To switch later, use the Subscription panel in your workspace. It shows what you will pay before you confirm.

• Monthly to yearly takes effect right away. You pay the yearly price minus a credit for the unused part of your current month, and your yearly term starts that day.
• Yearly to monthly takes effect at the end of your paid year. Nothing is charged until then, and there is no partial refund for the rest of the year. You can undo it before that date.

Switching is not available during a free period or while renewal is cancelled. For help, email support@pairrail.com.

Checkout for your PairRail subscription runs through our payment partner. PairRail does not store full card details. You can manage or cancel in the billing portal linked to your account, or email support@pairrail.com.

Every plan can email support@pairrail.com or use the support form. We are available Monday through Friday, 09:00 to 18:00 IST, and reply within one business day. Pro also gets a private Google Chat space with our team; request it from the Subscription panel in your workspace. Enterprise gets a private Slack or Google Chat channel, with response times set in your order form.

No. Enterprise is a custom contract for volume, SLA and multi-workspace needs. SSO and data residency are not available today. If you need them, they are scoped in your order form. Use Email us about Enterprise on the pricing page or email support@pairrail.com.

Agents can find you for free; only real work counts against your quota:

• Discovery is free. Basic catalog browsing (catalog_browse and /v1/public/catalog) is edge-cached and costs you nothing. Spiders and agents discovering your products never consume your quota.
• Quota queries. Dynamic search (catalog_discover), indicative pricing (pricing_calculate), readiness evaluations and authenticated agent operations count toward your plan's included quota, and only when they succeed.
• Never counted. Handshakes (initialize, tools/list, ping), static metadata, and health probes are permanently exempt. So are access requests (request_catalog_access), calls that fail, and retries that replay a quote you already have.

Pro includes 25,000 queries per month. Additional queries are billed at $2 per 1,000 queries ($0.002 per query) at billing cycle end.

Unauthenticated public calls never create overage charges and do not use your included 25,000 queries. They have their own monthly allowance of the same size; once it is used, anonymous traffic is paused with a 429. Only authenticated agents with approved machine keys use your included queries and overage capacity.

Autonomous AI agents, MCP clients, and LLM tool execution engines run in headless environments. They cannot click visual tiles or execute interactive browser scripts, so web CAPTCHAs would lock them out.

PairRail uses progressive access controls that work for machine clients:

• Per-client limits: unauthenticated pricing is limited to 30 requests per minute per client, with per-seller safeguards in each running instance. The hosting edge provides the global limit across instances.
• Free static discovery: Basic catalog browsing (catalog_browse) is edge-cached and unmetered to the seller.
• Optional proof-of-work: When a signing secret is configured and an anonymous compute client reaches its own limit, PairRail can return a scoped SHA-256 challenge. A solution never bypasses seller-wide or protected-route limits.
• Access-request filtering: Machine key requests validate contact-email syntax and block a maintained list of disposable domains. Seller approval is still required.
• No overage from anonymous traffic: public crawlers never create overage charges on your account.

Sandbox hard-caps at 1,000 queries/month. Upgrade to Pro to continue with 25,000 included queries plus soft overage for authenticated agents, or wait until the next monthly reset.

After you publish, open Agent gateway → Protocol rails and copy the MCP (and catalog) URLs. Those PairRail URLs include your sellerId, which is how Atlas routes agents to your workspace today. Paste the MCP URL into Claude, Cursor, or any MCP client. Custom domains on your own site are not required for Sandbox or Pro.

No for the working loop. Share stamped PairRail URLs from Protocol rails. If a crawler must hit yoursite.com/.well-known/ucp (RFC 8615 / Google UCP), open the same panel and copy the optional Next.js or Cloudflare Worker rewrite snippet, which proxies your apex to PairRail. PairRail does not manage your DNS or verify that proxy. Managed custom domains for protocol endpoints are an Enterprise option scoped per contract, not self-serve on Sandbox or Pro.

Discovery and indicative pricing can stay public so agents find and price you. Creating a firm quote or handing an approved quote to your billing requires a seller-approved machine key. Approve an inbox request or issue a key in Agent gateway; the secret is shown once. Limits, rotation and revocation are under agent access.

Every quote shows the agent key that asked for it, which you issued or approved. If the agent also signs its request with Web Bot Auth, Atlas checks the signature against the keys that agent publishes and shows you who signed and whether it verified. Signatures are recorded, not required, so they never block a request.

Sandbox serves MCP (plus REST /v1): discovery and indicative pricing are live, and quotes are marked as a Sandbox rehearsal. Pro and Enterprise unlock MCP, UCP, ACP, A2A and UAP with live quotes and no Sandbox marking. AP2, MPP and x402 remain on the roadmap. See Protocols.

On Sandbox, your catalog, discovery and indicative prices are live for agents. Quote and execute responses carry a Sandbox marking that says they are a rehearsal: the billing handoff runs only in Mock mode, never on your live billing. Move to Pro to send live quotes to your billing system.

No. Publish commercial truth once in Atlas. Protocol projections come from that catalog, so you are not maintaining a price book per rail.

Yes. The Protocols page opens live inspector views for the five commerce rails (/mcp, /.well-known/ucp, /protocols/acp, /a2a, /protocols/uap). The UCP product feed is part of UCP, not a separate rail.

Yes, as API documentation, not a sixth protocol rail. See /openapi.yaml for HTTP schemas used by the console and gateway. Commerce protocols remain MCP, UCP, ACP, A2A and UAP.

Not yet. x402, AP2 and MPP are on the protocol roadmap. PairRail can format x402 payment requirements from your published catalog in library code, but no payable x402 route is live, because payment verification and replay protection are not yet implemented and tested.

Paste or upload pricing content. Atlas structures products and offers, flags conflicts, and scores readiness. Deep verification reads long or messy sources section by section, checks every price against the source text, accounts for every amount the source prints, and flags anything it cannot verify so it cannot be published unchecked. Sandbox includes a limited personal AI extraction allowance and 3 Deep verification runs a month. Pro includes 100 Deep verification runs per month, plus the catalog copilot: change the draft in plain language, get options to fix a conflict, or sync the catalog with your pricing page, with every change shown before it is applied. Enterprise starts from a higher contracted monthly Deep verification allowance (1,000/mo in the base Enterprise ladder) and can raise volume further under the order form, with custom extract prompts and SIEM-exportable AI audit trails.

Sandbox is one seat. Pro includes 5 seats on the same seller workspace. The owner assigns each teammate a role: Owner, Publisher, Editor, or Viewer. Editors can draft and extract. Publishers can publish, pause live catalogs, and manage machine keys. Only the Owner invites people and manages billing. Enterprise adds multi-workspace and SSO role mapping under contract.

Both are Bearer keys (pat_…) with the same scopes, limits, expiry, rotation and revocation. They differ in who they are for:

• Internal machine keys are for your own agents and tools, such as a deal-desk bot, your own MCP client, or scripts that test quotes. You issue them in Agent gateway with Issue internal key and copy the key yourself. They can be allowed to see internal-only offers, and they never expire unless you set a date. Sandbox includes 1, Pro includes 5.
• Buyer-agent keys are for one outside buyer's agent each, such as a customer's procurement agent or a partner platform. The agent asks for access, you approve it under Access requests, and the agent collects its key once with a pickup token. They never see internal-only offers, and revoking one cuts off only that buyer. They expire 90 days after approval unless you renew them, and keys with no use for 30 days are flagged so you can revoke them in one step. Sandbox includes 10, Pro includes 250, separate from internal keys. Expired and revoked keys free their slot.

Browsing and indicative pricing need no key at all. A key is needed for firm quotes, and sending quotes to your billing stays off until you turn it on for a key.

You approve which machines can act on your catalog. Every plan includes scoped credentials with expiry, rotation, revocation, and per-key quote limits, so you can cut or throttle access without republishing. Per key you can set scopes (browse and quote by default; sending approved quotes to your billing only when you add quote:execute), an expiry date, maximum quotes per hour, maximum pricing calls per hour, a daily quoted total, and hourly and daily limits on handoffs to your billing. Each approved buyer gets its own key, which does not count against your internal key limit (Sandbox: 10 buyer keys, Pro: 250). Identity signals an agent sends with its access request (such as Web Bot Auth or Visa TAP tokens) are shown as claimed, not verified, so you decide how much to trust them.

Yes. Each offer has a disclosure setting: public, customer, or private. Public catalog, discover, and indicative pricing surfaces only show offers marked public. Customer offers stay gated for eligible buyers; private offers stay off public rails.

Yes. Publishers and Owners can pause a live catalog. That clears the live published pointer, keeps version history, and moves the Verified Agent-Ready mark to a paused state until you resume. Agents stop seeing a live catalog while paused.

Sandbox keeps about 14 days of activity. Pro retains audit trail and catalog versions for 90 days. Audit history is kept separately for each seller (up to your most recent 1,000 audit events), so other sellers’ activity never pushes yours out. Enterprise can extend retention under contract.

Core services run on Google Cloud in us-central1 (Firebase Authentication and Hosting, Cloud Run, and model extraction). Cloudflare stores catalog and workspace state (R2 and D1) and handles DNS, CDN and edge telemetry. The full list is on Subprocessors.

Stripe restricted keys, Chargebee API keys, webhook signing secrets and not-yet-claimed buyer agent keys are encrypted at rest with AES-256-GCM. Each value is bound to its organization or access request, so it cannot be decrypted in another tenant’s context. The encryption key is held in the runtime’s secret configuration, not in the data store, and the service refuses to store new secrets if no key is configured. Agent keys are stored only as SHA-256 hashes after the buyer picks them up. Atlas never stores customer payment card numbers. Customer-managed keys (Cloud KMS) are not available today. If an Enterprise contract needs them, they are scoped in your order form.

You do. PairRail processes Customer Content only as needed to operate the service (hosting, protocol projection, extraction, backups). See the Data Processing Terms and the Privacy Policy.

We do not sell Customer Content. Extraction uses configured model providers to transform content you submit for your workspace. Privacy explains how that processing works.

Email security@pairrail.com with “Security disclosure” in the subject, or follow the disclosure policy. How Atlas keeps agents inside your limits is on Security. support@pairrail.com still reaches us. For product misuse, see the Acceptable Use Policy.

Yes. Public processor terms are on the DPA page. Providers are listed under Subprocessors. Email support@pairrail.com if you need a signed DPA on an order form.