What a support triage agent actually costs, four ways

Every agent platform quotes you a monthly price. None of them quote you the price of your agent. So we took one small, boring, extremely common agent and costed it four ways, with the arithmetic in the open.

The agent. Support triage. It reads an inbound ticket, pulls three similar past tickets for context, drafts a reply and tags the thing. Nothing clever.

The month. Forty tickets a working day, twenty working days: 800 tickets. Three model calls per ticket. Roughly 3,000 tokens in and 400 out per call, so about 10,000 tokens a ticket and 8 million tokens for the month. Deliberately small. This is a two person support desk, not a call centre.

Every price below is what the vendor's own page said on 23 August 2026.

Lindy — $99.99 per user per month

Lindy's pricing page bands its credits by kind of work, and the example it gives happens to be ours exactly: "triage a day's support queue" sits in Deep work, 250 to 1,000 credits.

Take the floor of that band. 250 credits x 20 working days = 5,000 credits a month. Plus, at $29.99 per user per month, includes 3,000 credits, so it runs dry on day twelve. Pro, at $99.99 per user per month, includes 15,000 and covers the month comfortably. At the top of Lindy's own band, 1,000 credits a day, even Pro's 15,000 is gone by day fifteen.

There is no free tier, only a seven day trial.

Relevance AI — $234 a month

Relevance runs two meters: Actions, which are task runs, and Vendor Credits, which are the actual model spend at wholesale. Their pricing docs put Team at $234 a month billed annually, $349 billed monthly, including 7,000 Actions and $70 of Vendor Credits.

Our 800 tickets fit inside 7,000 Actions with room to spare. And 8 million tokens at wholesale rates — call it $0.30 per million in and $1.20 per million out — comes to roughly $3.30. Nowhere near the $70 allowance.

Which is worth saying plainly, because the usual complaint about Relevance is that the model costs are extra. At this volume they are extra and they are irrelevant. The subscription is the entire bill: $234 a month to run one small agent.

CrewAI — free until the second working day

CrewAI's pricing gives you 50 workflow executions a month on the free Basic tier. At forty tickets a day, execution number 50 lands mid morning on day two.

After that there is no named paid tier. The next step is Enterprise: custom pricing, request a trial, talk to sales. That may well be the right answer for a large deployment. It is not a number you can put in a spreadsheet before you build the thing.

Self-hosted — $0 in model fees

The fourth option is to run both halves yourself.

FreeLLMAPI is our MIT licensed router. It does not issue keys and there is nothing to sign up for. You collect free API keys directly from the providers who give them away — Google AI Studio, Groq, Cerebras, NVIDIA NIM, Cloudflare Workers AI, Mistral, none of which ask for a card — paste them into the dashboard, and the router puts all of them behind one OpenAI compatible endpoint at http://localhost:3001/v1. It handles routing, per key rate limits and failover to the next provider the moment a quota runs dry. As of August 2026 its README counts 29 providers, 251 model families and 358 free endpoints.

CircleChat is the other half: team chat where AI agents do real work. Self-hosted, it already treats custom:freeapi as its default agent provider. You point it at your router and the agents run on it:

OPENAI_BASE_URL=http://localhost:3001/v1
OPENAI_API_KEY=<your unified FreeLLMAPI key>

Eight million tokens a month sits comfortably inside the stacked free tiers. Model fees for the month: zero.

The honest costs

Zero in model fees is not zero. You are still paying for:

• A box. A small VPS, or a machine you already own. The chat stack fits in about a gigabyte of RAM and both halves run happily on a Raspberry Pi 4.
• Your time. Two services, a Postgres, a Redis, and the upkeep of all of it.
• Optionally $19 a year for FreeLLMAPI's signed live catalog feed, if you want the model list to keep itself current as providers add and retire models. The router is free and works without it.

And FreeLLMAPI's README is blunt about what stacked free tiers cannot do: no frontier models, variable latency, no SLA, and the effective intelligence of the endpoint dips late in the day as the best models hit their daily caps, then resets at UTC midnight. For support triage that is survivable. For anything where a wrong answer is expensive, it is not.

One more caveat, because it is the easiest thing to get wrong: the $0 applies to self-hosting both halves on your own hardware. CircleChat's cloud plans bundle a gateway with fair use limits, which is a perfectly reasonable trade, but it is a different bill.

So

$99.99, $234, an unpublished number, or the price of a small server.

The gap is not a scandal. Those platforms sell managed infrastructure, support, and someone to call at 3am, and for plenty of teams that is worth well over $234 a month. But you should know what you are buying with it, because at 800 tickets a month you are not buying tokens.

If you want to see the bottom of that range for yourself, install the router. It is one command, and every key it needs is free.