Relevance AI Review: Agents, Workforces and Two Meters to Budget

Relevance AI is a hosted low-code platform for building, running and managing AI agents and Workforces, its canvas for teams of agents that act inside business apps. Is it worth a look? For a team ready to design its own agents, yes. Self-serve pricing starts with Pro at $19 a month billed annually or $29 a month billed monthly, and that price covers the whole organization. Two meters drive the real bill: Actions and Vendor Credits. The Free plan is retired. Capability, AI, integrations and documentation are the strengths, while value and support below Enterprise hold it back.

What you actually build on Relevance AI

The platform has two faces: a builder for agents, tools, Workforces and knowledge, and Relevance Chat, where people use the agents. Agents are LLM-powered workers that receive tools and decide how to use them, either on autopilot or by asking a human first. Tools are chains of steps, such as LLM prompts, API calls, app actions and Python or JavaScript code, built without code and reused across agents. Knowledge adds retrieval over files, websites and synced Google Drive, SharePoint, Notion or Confluence^[p1] content.

A Workforce is where the platform parts ways with a single chatbot. Specialist agents sit on one canvas, with trigger, agent, tool and condition nodes joined by edges. Each edge either lets the agent decide when to hand over or always passes the task along, and each can require a human approval. Conditions route work by plain-language rules or by fixed comparisons, and since October 2026 they can also call dedicated decision models. The docs recommend Workforce over the older sub-agents feature for any multi-agent design.

Around that core sit the specialist modes. Phone agents place outbound calls in real time, and meeting agents join Google Meet, Zoom or Teams as a notetaker; both start on Team. Relevance Chat runs on the web, a desktop app and iOS and Android apps, but the mobile apps are for chatting only. Building happens in a desktop browser. Relevance AI is the trading name of OnSearch Pty Ltd, an Australian company, and the service runs hosted only, with no self-hosted edition described in its docs.

Two meters, one bill for the whole organization

Relevance AI's public pricing page shows only a custom Enterprise plan with a Talk to sales button; the self-serve Pro and Team prices are published in its pricing documentation, and you upgrade to them inside the app.^[1] Everyone in the organization shares one plan's allowances and features. Build Users and End Users are included counts, not priced seats, and no extra users can be added once a plan's limit is reached.

The first meter is Actions. One Action is a single run of a Tool, whether it sends one email or runs a many-step workflow inside that tool, and a run that fails still counts.^[2] Retries count too, and on Enterprise each Eval check costs one Action. Plan Actions reset to the plan amount at each renewal. The second meter is Vendor Credits, which pay for AI models and tools at wholesale cost with no markup, at $20 per 10,000. Plan and top-up Vendor Credits roll over indefinitely while you stay subscribed on Pro or Team.

Top-ups are packs at $80 per 1,000 Actions and $20 per 10,000 Vendor Credits, on paid plans only, bought by an organization admin. Purchased Actions roll over to the next billing cycle. That price matters: by our arithmetic, an extra Action costs $0.08, far above the included rate. Your own model keys are optional on every plan, but they cut only Vendor Credit spend, not the subscription or Actions.

By our arithmetic, Team at $349 a month billed monthly costs no more than Pro billed monthly plus Action packs above 6,500 Actions a month on monthly billing. On annual billing the crossover comes sooner, at about 62,000 Actions a year (roughly 5,200 a month). Both figures assume your model spend fits Pro's 10,000 Vendor Credits a month; heavier model use lowers the crossover. By our arithmetic, annual billing saves $120 a year on Pro and $1,380 a year on Team, and it costs less than monthly billing from the 8th month on Pro and from the 9th month on Team.

Three controls keep the meters in check. Usage alerts send an email at a level you set, and usage limits stop spending at a hard ceiling; both work on every plan. Spend Controls, on Pro and Team only, buy top-ups automatically when a balance drops below a threshold. Relevance AI takes credit cards only, not debit cards. An admin cancels in the app, the change takes effect at the end of the paid cycle, fees are not refunded for a change of mind, and unused Actions and Vendor Credits expire at cancellation. Our Relevance AI pricing guide works through the full cost scenarios.

Where Relevance AI is strongest

Invent, the built-in AI builder, is the quickest way in. You describe a business process, and it builds and coordinates the agents, tools, triggers and Workforces, then later proposes tested improvements. The safeguards are sensible. Invent works in draft by default, delete and publish actions default to asking every time, and nothing goes live until a person publishes it. The docs are candid about cost, warning that running Invent is costly and draws heavily on credits, and a session counter shows the spend as you go.

Reach comes next. Relevance AI connects to 2,000+ apps, combining its own native integrations with Pipedream-based ones, and offers 1,000+ app triggers on every plan. Premium triggers for WhatsApp, LinkedIn and Telegram come with every paid plan. Where no native connector exists, any system that can send an HTTP request can start an agent through a webhook, which is also the route for workflow tools without a native link. Zapier can trigger agents directly. Custom APIs plug in through an API tool step, so in-house systems are reachable too.

Model choice is open. You can let Relevance AI pick the model or choose OpenAI, Anthropic Claude, Azure OpenAI, Google Gemini or OpenRouter models, with a fallback model that retries a failed task once. For developers, the MCP server at mcp.relevanceai.com lets coding agents such as Claude Code, Cursor or Codex create, run and troubleshoot agents, and the documented SDK is JavaScript/TypeScript, published as @relevanceai/sdk. The documentation itself is deep, covering each integration, system limit and plan rule on its own page.

Control over live agents is built into daily operation. Each tool can run automatically, require approval or let the agent decide, and an autonomy limit caps how many tool runs happen in a row without a human. Every draft save and publish creates a restorable version of agents, tools and Workforces. Relevance AI says Qualified built 35+ AI agents with it and generated $7 million in pipeline in six months. Our guide to building Relevance AI agents walks through models, approvals, memory and triggers.

The limits you will meet first

Every organization now needs a paid plan, and Relevance AI's pricing pages describe no free trial. Evaluation costs money from day one. Because failed runs count, a buggy tool also burns Actions while you debug it. Pro stops early for teams. It includes two Build Users and no End Users, so colleagues who only use agents need Team. Calling and meeting agents, the analytics dashboard and bulk scheduling also start on Team. On monthly billing, that is the step from Pro at $29 a month to Team at $349 a month.

Governance sits higher still. SSO, RBAC, audit logs, agent evaluations, work-hour controls, model controls and Visual Data Masking (a limited beta) are Enterprise-only, as are triggers for Salesforce, Snowflake and Zendesk, even though their tool steps work on every plan. Concurrency is described only as Standard on Pro, More on Team and Custom on Enterprise, and tasks beyond the limit queue rather than fail. The API is officially supported only for triggering agents and tools, and that trigger returns no direct response. Our Relevance AI buyer's checklist turns these gates into questions.

Support below Enterprise runs on weekdays, 9am to 5pm Sydney time, with no weekend coverage, though engineers are on call around the clock for outages. First-response targets are three business days on Pro, two on Team and one on Enterprise. US teams should plan for the time difference. The data region, US, EU or Australia, is chosen when the organization is created and cannot be changed later. Pick it carefully. Moving regions means a new organization set up through support.

How our score reads Relevance AI

Our composite for Relevance AI is 4.0/5, built from eight weighted axes rather than from review averages. Capability, AI, integrations and documentation carry it: the Workforce canvas, Invent and the integration catalog are hard to fault on depth. Value pulls it down, because of the two meters, the jump from Pro to Team for calling, meeting and analytics features, and the absence of any unpaid way in. Support scores lower too, given business-hours coverage on Sydney time below Enterprise. Relevance AI sits mid-table among the AI agent builders we score.

Relevance AI has only a small number of public user reviews so far, so we score user sentiment as neutral rather than reading a trend into them. The neutral axis reflects too few reviews to read a trend, not a verdict on the product. The product is judged on what its documentation shows it can do. Teams comparing it with other agent builders should weigh the billing model as heavily as the feature list, since the same workload can cost very differently depending on how many Tool runs it needs and which model each step calls.

Is Relevance AI right for a team building its own agents?

Pick Relevance AI if your team wants agents that make decisions, hand work to each other under approval rules and reach into thousands of apps without anyone running servers. Start on Pro, watch how many Actions each task uses for a few weeks, and move to Team once you need calling or meeting agents or End Users. On Actions alone, Team pays off above about 6,500 a month on monthly billing, or about 62,000 a year (roughly 5,200 a month) on annual billing, by our arithmetic, if your model spend fits Pro's 10,000 Vendor Credits a month. Skip it if you need an unpaid start, self-hosting, weekend support or SSO without an Enterprise contract. Budget for both meters before you build anything.