Simple, call-based pricing
Know where model and MCP costs are quietly consuming margin.
Start with every feature for 14 days. See which customers, workflows, models, MCP servers, and tools are profitable before choosing a plan.
Start here
Full trial
Prove the economics on real production traffic before choosing a plan.
- All Growth features included
- Trial starts with your first ingested event
- No credit card required
- Unlimited customer IDs
For early-stage AI products
Starter
Know which customers, features, and workflows make—or lose—money.
- Track multiple applications in one workspace
- Unlimited customer IDs
- 90-day economics history
- Cost and profit by customer, feature, workflow, prompt, and execution
- MCP server, tool, and linked execution cost
- Versioned provider pricing and private rates
- Margin-risk alerts
For scaling AI products
Growth
Track margin movement over time and operate pricing with confidence.
- Everything in Starter
- 12-month economics history
- Cohorts, trends, drivers, and pricing headroom
- MCP fan-out and retry trends
- Revenue and billing-actual imports
- Historical repricing
- Provider-invoice reconciliation
For high-volume teams
Enterprise
Shape volume, retention, controls, and support around your operation.
- Everything in Growth
- Negotiated volume and retention
- Custom contract rate cards
- Custom model and MCP tool rate cards
- Provider-invoice reconciliation
- Dedicated implementation support
- Custom commercial terms
Why teams invest
Your AI bill cannot show the MCP chain behind the spend.
A provider can tell you that you spent $80. It cannot tell you that one MCP lookup for a customer paying $50 triggered planning, retrieval, validation, and three paid model calls.
- Subscription revenue
- $50
- Model + MCP-linked COGS
- −$80
- Contribution
- −$30
Without execution-level attribution, the MCP trigger and its downstream spend hide inside aggregate provider bills.
No metering games
You pay for economics processed—not raw agent exhaust.
Accepted model-call events
Each successful, failed, or retried provider call can carry real cost and therefore counts once when accepted.
Everything around the call
Duplicates, rejected events, revenue entries, repricing, dashboard reads, alerts, and configuration are not billable calls.
Customer attribution
Pass as many B2B tenant or account IDs as your product serves. The value of unit economics depends on seeing all of them.
Pricing questions
Know exactly what you are buying.
01Does Ganivra track MCP tools?
Yes. Send MCP steps through the same event API as model calls, identify the server and tool, and keep every step on the same execution ID. Ganivra can show the direct tool cost plus the planning, retrieval, validation, retry, and model costs the tool triggered.
02What is an accepted AI call?
One valid model-call event accepted by Ganivra for economics processing. Calls can come from OpenAI, Google Gemini, Anthropic Claude, Perplexity Sonar, or another provider through the REST event contract.
03What does not count toward usage?
Rejected events, duplicate event IDs, revenue-ledger entries, historical repricing, dashboard views, and pricing or customer-plan configuration do not count as AI calls.
04Do customer IDs increase the price?
No. Every paid plan supports unlimited customer IDs. A B2B AI company can send its existing tenant or account ID dynamically on each event and analyze every customer's cost-to-serve without registering them manually.
05Do we have to send revenue on every call?
No. Configure and assign a subscription, token, credit, per-execution, or hybrid customer plan in Ganivra. Your application sends the customer and usage facts—not prices. Credit plans send the units deducted; the other plan types use captured execution and token telemetry.
Fourteen days. Real traffic.
Find your loss-making AI and MCP workflows.
The trial clock starts only when Ganivra receives your first event.