Relevance AI

What it is

Relevance AI is a platform for building teams of AI agents, which it calls an AI workforce. Its builder is called Invent, introduced on the homepage as "Meet Invent 2.0, Your 24/7 AI solutions engineer." The page says Invent "helps your people identify what to automate, turn their expertise into reliable systems, and improve those systems over time." It is written for enterprise operations, sales, support and revenue teams that want agents handling real work, and for the engineers who support them. The footer names the company as OnSearch Pty Ltd, trading as Relevance AI. The pages do not state a location.

What you build and how

Relevance AI does not stop at a single chatbot. The unit it sells is an agent, and a "workforce" is several agents with tools and approval steps. The agents page describes three ways to build one, all on the same platform.

Drag and drop. The company says "Compose agents and workforces on a visual canvas. Drag in agents, tools, and approvals in a flow anyone on the team can read." The approvals matter: a person can sit in the flow where an agent should pause before acting.

Build with AI. "Describe the agent in plain language and Invent builds it: the prompt, the tools, and the evals to prove it works." The example on the page is an agent that triages support tickets. The demo shows Invent writing the system prompt, building 3 tools from your docs and writing 24 evals from past runs, ending with a "Ticket Triager" agent. You still review the result, but you are not starting from an empty prompt box.

Build with MCP. "Engineers drive the same platform from Claude Code, Codex, or Cursor. Create agents, link knowledge, and run evals over MCP." This is the route for a developer who would rather script the platform than click through it.

Triggers and connectors. The agents page says agents receive tasks through "native triggers for CRMs, email, calendar, and more," plus webhooks, cron, and the ability to build custom triggers in the platform. For access to your systems it states "1,000+ native connectors plus MCP support and the ability to build custom connectors." On cost, the company says you can "route to any model provider and optimize cost per agent, per task." Its pages do not list which connectors or which model providers.

Who runs it day to day. The company's answer is "your ops and revenue teams, the people closest to the work", with "dashboards and tooling designed specifically for agent management." Quality is owned by "domain experts, with pre-deployment scenarios, production checks, and monitoring they can manage themselves." The page also says there is lots of control over approvals and escalations, and a pane for access control, permissions and data security.

What agents do. The homepage shows sales agents named Outbound Prospector (multi-channel prospect engagement), Forecast Roll-up (rolling deal signals into a pipeline call), Deal Reviewer (checking deals against criteria and flagging risk) and Proposal Builder (drafting proposals and quotes). Other departments listed are customer success, marketing, human resources, support and operations, with support triage and content marketing named as examples. A "Contract Reviewer" and an "Invoice Matcher" appear on the rollout timeline.

Autonomy levels. The homepage describes four stages: L1 Assisted (delegate research and drafting to a copilot), L2 Copilot (teach it your playbooks, "powerful, but hard to govern"), L3 Autopilot (proven playbooks become governed agents that run autonomously) and L4 Self-Driving (your agents run evals and swap models themselves). Teams are encouraged to build "Level 3 and Level 4 autonomous agent workforces" by week 4 onward.

The pages do not describe a publishing flow to an app store, a code-export option, or self-hosting.

The evidence from the pages

Pricing. No plan, price or credit allowance appears on the pages. Calls to action read "Book a demo", "Talk to sales" and "Try for free". The comparison page says "We partner with enterprise teams to build agents proven against your quality bar and live in weeks. Bring one workflow and compare for yourself."

Rollout. The homepage lays out a timeline. Weeks 1-2 map initial use cases and value. From week 3, you deploy a first team of agents with Invent, and "some use cases can take as little as a few hours to stand up, with evaluation, monitoring, and governance." From week 4, you evaluate the pilot, make it production ready and start building new use cases independently. It adds that "an embedded deployment team gets you live, then trains your team to build quickly on its own."

Customer results. These are the company's own case-study claims. Qualified "generated $7M pipeline with 35+ AI agents across the org", alongside a 10x increase in output and $7M of pipeline in 6 months. Send Payments "saves 40 hours weekly by going AI-first across global operations". Zembl "drives 30% more customer conversions with 24/7 AI sales coverage", with a 60% faster average call time. The company also says it has "helped thousands of teams launch their AI workforce" and that it puts agents to work on "millions of tasks every month".

Security and compliance. Under "Built for enterprise teams" the homepage promises "access controls, audit trails, and residency guarantees", and shows SOC 2 and GDPR badges next to a security and data privacy list that includes data residency and PII masking. The pages do not give a report date or scope.

Comparison with Gumloop. The company's comparison page calls Gumloop "a no-code platform for building AI agents" where business users build self-serve, and says "Both are good tools." It positions Relevance AI as wiring agents "deep into your systems" with a team alongside you.

Who it suits and who it does not

Relevance AI suits a company that wants several agents across departments and wants help getting the first ones live. A revenue operations lead can build an outbound or deal-review agent, an ops team can own the evals, and an engineer can still reach the same agents through MCP. The structured rollout, with weeks of mapping before deployment, favors teams that would rather prove one workflow than experiment alone.

The caveats follow from that. There is no published price, so budgeting means a sales conversation. The pitch leans on an embedded team and a multi-week rollout, which is more than a founder wanting a website chatbot this week needs. Nothing on the pages promises a quick self-serve chat widget, an embeddable bot template or a documented free plan. And because the compliance badges appear without detail, a regulated buyer should ask for the reports before relying on them.