Platform pricing
Priced per AI step.
Meridian charges for the work agents do. One monthly fee covers the platform and a volume of AI steps; beyond it you pay per step. Run the models through us or on your own provider account.
a month for the platform, with 375,000 AI steps included
per 1,000 AI steps beyond the included volume
AI usage at cost + 20% through us, or billed by your own provider
Calculator
Calculate your monthly cost.
Tell us how many tasks it handles a month and how much work each one takes. The estimate uses the step counts and token profiles we measure in production.
Tickets, conversations, leads, documents: whatever it handles one at a time.
Several steps: lookups in your systems, a search, a drafted reply.
Billed by us at the provider's price plus 20%, on the tokens used.
$0.0064 of AI usage per step.
Run the booking, the write or the decision on a stronger model; the rest stays on the base model.
| Per month | Best | Forecast | Worst |
|---|---|---|---|
| AI steps | 150,000 | 225,000 | 325,000 |
| LLM | $960 | $1,440 | $2,080 |
| Platform | $1,500 | $1,500 | $1,500 |
| Per task | $0.10 | $0.12 | $0.14 |
| Total | $2,460 | $2,940 | $3,580 |
A standard step costs $0.0046 of AI usage on DeepSeek V4.1 Flash, measured on knowledge-retrieval threads, 169 threads, blended with the short turns around them. Model prices from the Vercel AI Gateway catalog, 4 Nov 2026. Steps per task are blended across how far the work gets, from a request that bounces to one that runs to the end, and measured on our own workspace's usage (30 days to 3 Nov 2026, outliers trimmed) and on live customer deployments. The estimate is a forecast; AI usage is billed on actual use.
What you pay for.
One AI step is one model call.
Every time an agent reads its context and decides what to do next, that is a step: answer, search, look up a record, call a tool, draft a reply. A simple question takes a handful; a booking with writes takes twenty. Step counts and cost per step come from the agents running in our own workspace and in customer deployments.
- Reading the context and deciding the next action
- A search, a lookup or a tool call
- Drafting the reply or the record change
Your models or ours.
Agents run on the model you pick, and a skill can switch to a stronger model for the one step that needs it. Through Untethered, AI usage is billed at the provider's price plus 20%, which covers the gateway, retries and observability. On your own provider account, the models bill you directly and only the platform fee is on our invoice.
- Frontier, open-weight and on-premise models
- A stronger model for the critical step only
- Bring your own keys, or run through us
Where the platform fee sits.
Strategy and deployment are one-time fees, quoted per engagement. The platform runs on usage from the day the first agent is live.
Step 1
Strategy
- Two hours with your leadership
- Your operations mapped, ranked by impact
- A business case and KPI targets
- A proposal: scope, team and timeline
Step 2
Deployment
- An engineer and a product lead on site
- Your systems connected, context built
- A working agent in production
- Your team alongside from day one
Step 3
Platform
- 375,000 AI steps a month included
- Then $4 per 1,000 steps
- AI usage at cost + 20%, or your own
- No seats, no tiers
Questions about pricing.
A seat prices a person; a step prices the work. Agents that handle a workflow end to end have no seat, and a per-step price grows with what they do for you, not with how many people log in.
Next step
Put a number on your first agent.
Two hours with your leadership. Ten days to a business case. Four weeks to a working agent on your data. No commitment.