Relevance AI

Relevance AI keeps two official price views. The pricing page on the main site lists a single Enterprise plan behind “Talk to sales”. The documentation pricing page lists Pro from $19 a month and Team from $234 a month on annual billing ($29 and $349 billed monthly), with “Get started” links into the app, and its FAQ says you upgrade from the plan counter at the bottom left of the dashboard. The same page says the Free plan is retired for new signups and existing accounts, so every Organization now needs a paid plan.

An Action is one run of a Tool. The plans and credits page says it counts the same whether the tool sends one email or runs a complex workflow with many steps, and a tool that fails still counts. Vendor Credits are the model and tool costs, passed through at wholesale with no markup; at the top-up price of $20 per 10,000, one Vendor Credit is $0.002. Bring your own model keys and Vendor Credits are bypassed, but Actions are still charged.

That makes design a cost decision. Steps you always run together are cheaper as one tool than as several tools an agent picks between. The failure policy in the agent tools docs can retry an errored tool up to a set number of times, and the docs warn that more retries may raise usage; each retry is another tool run. If you want run costs broken into model, tool-call and compute credits instead, the Gumloop profile explains that meter.

The tools docs describe a tool as inputs, steps and outputs built in a no-code builder: LLM prompts, API calls, integrations and Python code. An agent decides when to use which tool, and a workforce links several agents and tools on a canvas with conditions between them. Triggers come from app events, schedules, webhooks or API calls.

Approvals are set per tool or per connection. Approval Required makes the agent draft its intended action and wait; Let Agent Decide leaves the call to the agent; Auto Run can be capped with “Max auto runs”. A task waiting for approval stays open until someone approves or rejects it. Runs are limited to 15 minutes for most tools and for each agent turn, and tasks queue when your tier’s concurrency is full.

Three settings cover it. Usage limits, available on all plans, hard-stop Actions or credits at a monthly amount for the Organization or a project and email the people you list; usage alerts only notify. Spend controls on Pro and Team top up automatically, and the amount is easy to misread: with 1,000 credits left, a 10,000 threshold and a 1,000 top-up, you are charged for 10,000 credits to reach 11,000. If a payment fails, the account may be limited until it is fixed.

The security overview says Relevance AI is SOC 2 Type II and GDPR compliant, encrypts data with TLS 1.2+ in transit and AES-256 at rest, and stores it in the US, EU (London) or Australia (Sydney) as chosen at signup. It says your data is not used to train models or improve the services without a partnership agreement, and deletion requests are processed within 60 days. The terms, however, grant a licence to use Client Data to provide and to improve the Solution, so ask which wording your contract follows. SSO, RBAC and audit logs are Enterprise features.

1. Subscribe to Pro monthly, choose your data region, and set a usage limit of 300 Actions for the test project.
2. Build one tool with three steps (classify a lead with a prompt, add a sheet row, post to a test Slack channel) and run it on ten sample leads.
3. Build an agent that calls three single-step tools for the same job, set the Slack tool to Approval Required, and run the same ten leads.
4. Reject one approval and break the sheet connection once, then read the Actions and Vendor Credits each run used.

Accept if the tool build uses about ten Actions, the agent waits on every Slack post, and the rejected post never appears. Reject if retries consume Actions you cannot see in the breakdown, or the agent posts without approval. No result is recorded here.