Relevance AI: AI agents billed per action, pricing and the end of the free plan

What Relevance AI does

Relevance AI is a low-code and no-code platform (little or no code to write) for building AI agents. An agent gets an instruction, then picks for itself among “Tools” you give it: send an email, read a CRM record, look something up. Several specialized agents can work together; the vendor calls those teams “Workforces”.

The vendor is OnSearch Pty Ltd, an Australian company. Agents are triggered from messaging tools and apps: Gmail, Outlook, Slack, WhatsApp Business, LinkedIn, Telegram (triggers); calls and meetings from Team up source. The pages read mention no chat widget to put on a website (Relevance AI documentation, pricing page (accessed November 11, 2026)): this is not a customer service tool you install and switch on.

If agents are new to you, read the guide to AI agents.

The billing unit: actions, plus Vendor Credits

So there are two meters: actions, included in the subscription, and Vendor Credits, which pay for language models. The Pro plan costs $29 per month ($19 per month billed annually) and includes 2,500 actions per month, for the whole organization.

Example: an agent that handles inbound requests

Take an agent that runs four tools for each request it receives: find the contact in the CRM, read the history, send the reply, update the record.

• Four tools per request make 4 actions. With 2,500 actions, the Pro plan covers 625 requests a month.
• Under the rule, a tool counts as one action even when it chains several steps. If the four steps are packaged into a single tool, each request costs one action: 2,500 requests a month.
• A run that fails still counts. A tool that fails one time in ten, and that the agent runs again, adds about 10% more actions for that tool.

How you slice your tools weighs on the meter as much as the number of requests.

When the quota is not enough, top-ups can be bought without changing plans, in blocks of 1,000 actions or 10,000 Vendor Credits (prices are in the box above). Plan actions reset at each renewal; topped-up actions carry over to the next cycle. The documentation does not say what happens to agents when actions or credits run out with no top-up (Relevance AI documentation, pricing page (accessed November 11, 2026)).

The site’s cost calculator covers automation tools, n8n among them, not agent platforms: for Relevance AI, the math is done by hand, as above.

The plans in brief

What has changed compared with what you may still read elsewhere:

• there is no free plan anymore. It is retired for new signups and existing accounts alike, and no free trial is mentioned;
• the website’s pricing page shows no prices. It shows only an Enterprise plan on quote; Pro and Team prices come from the official documentation;
• the pricing model dates from September 8, 2025. The former “Business” plan was removed and its customers were moved to Team.

Between Pro and Team, the gap is not mainly about volume. Team includes 2.8 times the actions of Pro (7,000 versus 2,500), while the price gap is far wider: compare the two rows of the plan table further down. What Team adds is features: calling and meeting agents, an analytics dashboard, 5 “Build Users”, 45 “End Users”, priority support.

The vendor advertises 33% off the monthly rate with annual billing. Subscriptions are paid by credit card only; debit cards are not accepted.

Hosting and data

The region is chosen at signup and cannot be changed afterward; conversations and the knowledge base are stored there. The record says: Chosen at signup: "US (N. Virginia)", "EU (London)" or "AU (Sydney)"; the "EU" region is London source.

The vendor advertises SOC 2 Type II and GDPR on all plans. The documentation also publishes technical limits: a tool run lasts up to 15 minutes or 24 hours depending on its type, and concurrent tasks beyond the plan’s ceiling are queued.

AI, models and MCP

• Models: OpenAI (GPT) and Anthropic (Claude) by default; Gemini, Azure, OpenRouter; your own API keys on all plans.
• Knowledge base: the “Knowledge” feature takes files, Google Drive, SharePoint, Notion and website content.
• Integrations: "2,000+ Apps" and "1,000+ App Triggers" on all plans, including the Pipedream catalog.
• MCP: this protocol, which connects an AI assistant to outside tools, is listed by the vendor among the integration methods. The principle is explained in the MCP guide.

Who it suits, and the alternatives

Relevance AI suits people who want to design their agents and are willing to configure the agents’ tools themselves. Depending on your situation:

• you want an assistant to connect to Slack and email rather than a platform for building agents: see Lindy, billed in credits per user;
• you want to answer customers on a website: see Tidio or Botpress;
• you want to pay per full workflow run, whatever the number of steps, or host the tool yourself: see n8n and the guide to building an AI agent.

All five tools sit on the same rows in the AI agent builder comparison.