Hey Product Hunt! 👋 I'm Naazim, and I built Tollgate which is a real-time gross-margin observability for AI agents. The problem I kept running into: AI companies charge per seat, per resolution, or per run. But their cost is non-linear token usage that swings wildly by customer and workload. A blended margin can look perfectly healthy while individual customers quietly destroy your unit economics. Here's the scenario that haunted me: Two customers, same $0.50 per resolved ticket plan: Customer A • Workload: FAQ bot (short, cheap) • Cost per ticket: ~$0.02 • Gross margin: +96% 🟢 Customer B • Workload: Billing disputes (heavy reasoning + tools) • Cost per ticket: ~$0.53 • Gross margin: -6% 🔴 Blended together, they look healthy (~40%). The number you're watching doesn't exist in reality—it's hiding Customer B, who is actively costing you money on every ticket. Tollgate breaks the blend apart and flags Customer B. What Tollgate does: Tollgate sits on top of your billing system (Stripe/Orb). It does not invoice your customers. It answers one question: "Do the prices you're already charging actually clear your AI provider costs?" Core capabilities: • Overview dashboard Blended gross margin, recognized revenue, provider cost, unprofitable customer count, revenue vs. cost trends, cost breakdowns, and a Margin Leaks table highlighting your worst-performing accounts. • Customer-level drill-down Searchable customer roster with margin badges, pricing model filters, and detailed breakdowns showing per-agent costs, token usage, and multimodal usage. • Margin Simulator Replay historical workloads against hypothetical pricing plans or AI models. Compare baseline vs. simulated margins before making pricing decisions. • Logs Complete audit trail of every AI run with revenue, cost, and margin. Includes detailed token breakdowns for input, output, reasoning, cached tokens, audio, images, video, web search, and tool calls. • Alerts Real-time alerts for: * Negative Margin * Reasoning Token Spike * Low Margin Threshold * Cost-per-Run Spike Alerts support email or webhooks with automatic retries. • Six pricing models Per-unit, per-resolution, usage-based, per-seat, flat monthly, and hybrid pricing. • Cost engine correctness Reasoning tokens bill at output rates. Tollgate separately accounts for input, output, reasoning, cached tokens, cache writes (5-minute and 1-hour TTL), audio, image, video, web search, and tool calls. That's the difference between showing a 96% margin and accidentally reporting a -6% customer as profitable. Provider pricing automatically syncs daily. • Zero-friction SDKs Wrap your existing Anthropic, OpenAI, Bedrock, or Gemini client with one line. Published as: • npm: @tollgateai/sdk • PyPI: tollgateai Zero dependencies. Only token counts and metadata are sent, never prompts or completion content. TypeScript quickstart ```ts import Anthropic from '@anthropic-ai/sdk'; import { createTollgateClient, wrapAnthropic } from '@tollgateai/sdk'; const tollgate = createTollgateClient({ apiKey: process.env.TOLLGATE_API_KEY, }); const anthropic = wrapAnthropic(new Anthropic(), tollgate, { customerId: 'customer-b', revenueUnitCents: 50, }); await anthropic.messages.create({ model: 'claude-sonnet-4-6', max_tokens: 512, messages: [ { role: 'user', content: 'Resolve this billing dispute…', }, ], }); ``` Links Live demo https://www.tollgateai.dev Video demo https://youtu.be/Rk8MRTQAMeY npm SDK https://www.npmjs.com/package/@t... PyPI SDK https://pypi.org/project/tollgat... Who this is for If you're building AI products, customer support agents, coding assistants, document processors, research agents, or any AI workflow where you charge by outcome, seat, or usage, Tollgate shows whether each customer is actually profitable. I'd love to hear your feedback. What metrics or integrations would be most valuable for your AI business?

Tollgate

Real-time gross margin for every AI agent run

Real-time gross margin for every AI agent run