Plans, Tokens, and Paddle Billing

SmartGate ships with four public plans. Here's how tokens and billing fit together.

Plan snapshot

All plans include full MCP tool access; limits scale with tokens, API keys, webhooks, retention, and fair-use RPM / plan limits (Free is intentionally strict).

Paddle as Merchant of Record

Paid subscriptions checkout through Paddle. They handle tax, invoicing, and payment collection where applicable. Manage upgrades, downgrades, and cancellation from Dashboard → Settings → Billing or the Paddle customer portal.

See Pricing for current prices.

Free tier

Free includes full MCP tools with strict fair-use rate limits (soft-fail HTTP 429 under burst). It is for individual evaluation; Playground is not for multi-user load. Over-limit bursts may receive HTTP 429. See /pricing for RPM and plan limits.

Start with 2M tokens/month — enough to wire up MCP and run real agent tasks before you upgrade.

Get started free

What you are buying

You pay for platform access to an intelligence layer and for governed tool traffic—not for SmartGate to host your chat model. Host LLM invoices stay with your provider. Dashboard Est. USD saved uses an internal gateway benchmark so you can compare tool paths and understand Gainshare inputs.

Gainshare without mystique

Where savings share applies, you pay a platform fee first. A variable share starts only after a free savings allowance. Search-class estimated savings may count at a reduced weight; fetch, compress, dedup, pipe, and budget-block savings count fully per published rules. Memory and Playground-only counterfactuals stay out of the share tax base. See /pricing for ceilings and FAQ detail.

Entitlements that change operations

Paddle specifics

Paddle is Merchant of Record: tax, invoicing, and payment collection follow Paddle policies. Manage subscription changes from Dashboard billing settings or the Paddle portal. We receive status and plan metadata to provision accounts—not full card numbers.

Choosing a tier

• Free — wire MCP, run real but bounded tasks, accept strict fair-use 429s.
• Pro — higher pool, longer logs, Gainshare with published ceiling, optional hard cap.
• Teams — shared workspace limits and governance for multiple builders.
• Enterprise — contractual pools and process; contact sales rather than extrapolating public tiers.

Positioning reminders for finance reviews

SmartGate is not a ChatGPT alternative or model router. Do not paste gateway counterfactual savings into the GL as if they were vendor credits. Do not invent ROI case studies. Pair this post with Billing docs, Security, and Contact for procurement questions.

The four tiers, by the numbers

Plan limits come from two places in the repo: the catalog in config/plan-catalog.ts and the rate-limit defaults in backend/smartgate/core/plan_entitlements.py. Read together they give the ceilings below; treat them as catalog defaults — /pricing is authoritative.

Monthly token caps are the monthlyTokenLimit values in config/plan-catalog.ts (2,000,000 / 20,000,000 / 100,000,000 / 200,000,000); retention days, API-key counts, and the REST-write figures are the PlanFeaturesV1 entries there, and the MCP numbers match CATALOG_RATE_LIMITS in plan_entitlements.py. Enterprise's 9999 webhook and key values stand in for "by contract" — not a literal count.

Two MCP limits, one team ceiling

Two different MCP limits apply at once, and confusing them is the usual source of "it works for me, not for us". mcp_rpm_per_key caps a single API key; mcp_rpm_team_ceiling caps the whole team's MCP traffic. On Free both are 30, so minting a second key buys nothing — the team ceiling is already saturated by one key. On Pro the pair is 300 per key against a 600 team ceiling; on Teams, 600 against 3,000; on Enterprise, 1,200 against 9,999. Playground has a separate rate (capabilities.maxPlaygroundRpm: 10 / 60 / 120 / 300 by tier), so evaluation load does not consume your MCP budget.

plan_entitlements.py also documents the safety behavior: resolved limits are cached for 60 seconds and fail safe to Free defaults on lookup errors, so a database blip degrades to strict limits. Per-team features can override those defaults via merge_rate_limits — which is why a negotiated Enterprise contract can exceed the numbers above.

Platform fee and the share ceiling

The paid tiers are a platform fee plus a capped sharing mechanism, and the exact numbers live in config/billing-savings-share.ts (PRICING_VERSION "2026-06-b4-balanced-v2"). Pro has a base fee of $18/month, a free savings allowance of $15, a variable coefficient (tau) of 0.10, a variable-fee cap of $18, and a maximum total fee of $36/month. Teams has a base fee of $55/month, a $35 allowance, tau 0.08, a variable cap of $45, and a $100/month maximum total. The variable component starts only after the allowance is consumed and is itself capped.

Not all savings count equally toward the share. The search-class multiplier SEARCH_BILLING_KAPPA defaults to 0.3, so search-class estimated savings count at 30% of the $2.50/M benchmark, while direct-tier tools (fetch, compress, dedup, pipe, and budget block) count at 100% (billing-savings-share.ts, formatSearchKappaFieldHint). Pro also carries an introductory label, "Try Pro — $5 first month". As with every tier, confirm the live figures on /pricing before quoting them in a budget line.

Team habits, the pre-ship checklist, product links, and the closing note live once for the series — see What Is an MCP Gateway?.