Relevance AI

What it is

Relevance AI is a no-code platform for building and running teams of AI agents — what it calls an “AI workforce” — over the tools an ops team already runs. You assemble agents in a visual builder by writing natural-language instructions, picking an underlying model (Claude, Gemini, or OpenAI), and granting each agent a set of actions it is allowed to take. Agents hand off to each other on a Workforce canvas: a BDR job is really a research agent, a scoring agent, and an outreach agent under one playbook, wired either by an AI connection (the agent decides when to pass control) or a fixed next-step transition. The platform markets a four-rung autonomy ladder — Assisted, Copilot, Autopilot, Self-Driving — so a process can start with a human approving every step and graduate once you trust it. It connects to 2,000+ apps, fires from 1,000+ app triggers, and speaks MCP in both directions: a hosted MCP server for Claude Desktop, Cursor, VS Code and ChatGPT, a Claude Code plugin, and an MCP client that lets your agents call external servers. The company is Sydney-founded and raised a $24M Series B led by Bessemer Venture Partners in May 2025, about $37M total.

Why it shows up in GTM stacks

The builder is general-purpose, but the center of gravity is GTM, and the finished agents Relevance markets are sales-shaped. Bosh, the flagship BDR agent, researches leads, enriches contacts, scores against ICP, writes outreach, handles replies, books meetings, and updates the CRM — it is itself a multi-agent system, and it is Enterprise-only, custom-onboarded rather than switched on. The rest of the catalog is task-shaped rather than persona-shaped: pre-meeting prepper, post-call actioner, outbound prospector, forecast roll-up, deal reviewer, proposal builder, contract reviewer, invoice matcher.

Inventor, added in 2026, is a copilot inside the UI that builds, debugs, and version-rolls-back agents from a plain-language description, using the platform’s own MCP to act on your project. It is an Enterprise-plan pilot; lower tiers get a lighter “invent an agent” prompt in the create flow rather than the full debugging loop.

Outside sales, teams build their own — a legal-ops intake triager that routes contract requests, or a recruiting agent that summarizes a candidate against the scorecard. Those are builds, not templates you switch on. Because every agent runs through the same connector graph, model layer, and approval gates, a RevOps agent and a recruiting agent are governed under one policy instead of two disconnected point tools.

Pricing

One structural thing to know first: the marketing pricing page now shows a single Enterprise “talk to sales” card. The self-serve ladder was not withdrawn — it is published in full in the docs, which is where the real numbers live.

Two meters, not seats: Actions (each step an agent takes) and Vendor Credits (the model cost behind those steps).

• Free — $0. 200 Actions/month, 1,000 Vendor Credits one-time, 1 Build User, 1 Workforce, 30-day task history.
• Pro — $19/month annual, $29/month monthly. 2,500 Actions and 10,000 Vendor Credits (~$20) per month, 2 Build Users, unlimited Workforces, scheduled triggers, escalations, bring-your-own-LLM.
• Team — $234/month annual, $349/month monthly. 7,000 Actions and 35,000 Vendor Credits (~$70) per month, 5 Build Users plus 45 End Users, calling and meeting agents, A/B testing, analytics.
• Enterprise — custom, no published price. This is where Bosh, Inventor, agent evaluations, work-hour controls, SSO/RBAC/audit logs, and the Salesforce, Snowflake, and Zendesk triggers live.

Annual billing saves 33% over monthly. The number to actually model is the top-up rate, because a production multi-agent process exhausts its included allowance long before it runs out of seats: $80 per 1,000 extra Actions and $20 per 10,000 extra Vendor Credits, bought in those increments on any paid plan. On Pro and above, connect your own Anthropic or OpenAI key to pay provider rates directly and drop the credit meter out of the equation entirely.

Agents, tools, and integrations are unlimited on every tier. Build Users are not — 1, 2, 5, unlimited.

Best for

• RevOps and GTM teams standing up multi-step agent processes — research, score, outreach, follow-up — behind human approval gates, with a named owner for agent design. ROI is best on Team, where calling and meeting agents plus the 45 End User pool let one or two builders serve a whole revenue org.
• Ops teams in any function that want a no-code builder over an engineering project, and that want model choice and MCP-native tool calling rather than a single hardwired model.

Not for

Do not buy this if your agents need to fire off CRM events. Salesforce, Snowflake, and Zendesk triggers are Enterprise-only — on Pro or Team you are scheduling agents or triggering them from WhatsApp, LinkedIn, Telegram, and webhooks, and the Salesforce integration is something your agent calls out to, not something that wakes it up. A 10-person RevOps team wanting closed-loop CRM automation on a $234/month budget will not get it here; that team is better served by n8n plus its own CRM webhooks.

Alternatives and when to pick them

• Microsoft Copilot Studio — pick when your team lives in Microsoft 365 and you want agents inside that estate with the largest default install base, not a separate platform.
• Lindy — the fastest-growing no-code agent entrant buyers weigh against Relevance, and it now prices per user with pooled credits. Pick it for lighter individual-to-team automations; Relevance pulls ahead on multi-agent handoffs and enterprise governance.
• Dust — pick when the job is a knowledge-grounded assistant over your SaaS (“where’s our position on uncapped indemnity?”) more than a process that takes actions. Dust leads on retrieval and flat published pricing.
• n8n — pick when you want to own the orchestration and wire models into a DIY workflow rather than buy a managed agent layer, or when the CRM triggers you need are Enterprise-gated here.

Watch-outs

• Actions plus Vendor Credits are the real budget, not the $19. A chatty multi-agent process burns Actions fast, and model credits scale with retrieval and tool calls. Guard: run a 30-day pilot on your single highest-volume use case, read Action and credit consumption off the dashboard before signing anything annual, and connect your own model keys to pay provider rates.
• The features that make it an enterprise platform are all on the tier with no published price. Bosh, Inventor, agent evaluations, SSO/RBAC, and the CRM triggers are Enterprise-only, and the marketing pricing page no longer quotes a number for it. Guard: get the Enterprise Action and Vendor Credit allocation in writing against your piloted consumption rate, not as “custom” — the top-up rates are your ceiling if the allocation is short.
• The marquee agents are GTM; the “AI workforce” claim is broader than what ships finished. For a legal-ops or recruiting process you are assembling agents in the builder. Guard: scope a build sprint with a named owner before assuming an off-sales process ships in a week.
• Autonomy without governance is silent bad output at scale. A Self-Driving agent that writes to Salesforce or sends email can degrade deliverability or data quality before anyone notices. Guard: keep write-capable agents on the Assisted/Copilot rungs behind approval until reply-rate and accuracy metrics hold, and promote one process at a time rather than flipping the whole workforce at once.