Relevance AI

Where Relevance AI sits

Relevance AI is an Australian-founded agent platform — launched in 2020 by Jacky Koh and Daniel Vassilev, high school friends with machine learning and engineering backgrounds — that has repositioned itself around a single idea: an "AI Workforce". The company raised roughly $15 million in a Series A led by King River Capital with Peak XV's Surge, Galileo Ventures, and Insight Partners participating (the sources split on whether that is AUD or USD — Forbes AU and SmartCompany report "$15 million", inaitoday reports AU$15M), then a $24 million Series B that Forbes AU reported alongside a total funding figure of $37 million. The co-CEOs relocated to San Francisco in 2025 to establish a US headquarters, and the marketing pages now name-check enterprise deployments at Canva, Autodesk, KPMG, and Lightspeed.

The pitch is straightforward: business teams describe the work, the platform builds specialist agents — sales qualification, support triage, research, content — and a "Workforce" layer coordinates those agents into running processes end to end. It sits in the crowded no-code agent category alongside very different neighbours: visual builders like Flowise for developers, and automation platforms like Zapier for workflow people. Relevance's bet is that the buyer is a business operator who wants staff-shaped agents, not flowcharts.

What it actually does

The building blocks, per the vendor's docs and pricing page, are agents, tools, knowledge, and workforces. You create agents by describing them in natural language and then refine behaviour, knowledge sources, and tool access in a visual builder; the platform claims 2,000-plus integrations, and premium connectors (LinkedIn, WhatsApp and similar) appear from the Team tier upward. Tools are the unit of work — each tool run is what the platform bills as an "Action" — and agents chain tools in multi-step runs, with execution logs exposed in a dashboard so you can see what an agent actually did. Workforces orchestrate multiple agents against a process, with scheduling and ticketing hooks in the mid tiers, and the higher tiers add agents that make phone calls, agents that join meetings, A/B testing of agent behaviour, an analytics dashboard, higher concurrency, and larger knowledge stores. The documentation describes the goal plainly: agents that complete tasks "much like" human workers.

Billing has two currencies, and this is the part to understand before you commit. Plans bundle Actions — counted every time an agent runs a tool — and separately pass through Vendor Credits, the underlying LLM and tool costs, which is where usage cost lands when agents do heavy model work. Third-party analysis (deciderstack.com) also reports a bring-your-own-key option for routing model costs through your own accounts; I could not confirm that on a public vendor page, so treat it as reported rather than promised.

Pricing at time of writing, from relevanceai.com/pricing-new in October 2026: Free at $0 with 200 Actions per month plus a small bonus of vendor credits, unlimited agents and tools, one workforce, one user, one project, and seven-day task history. Pro at $29 month-to-month or $19 per month billed annually ($348 per year at full price before the annual discount). Team at $349 month-to-month or $234 per month billed annually ($4,188 per year), with 30,000 Actions per month (35,000 billed annually), 50 users, five shared projects, vendor credits worth $70, agents that make calls and join meetings, A/B testing, analytics, higher concurrency, and priority support. Enterprise is custom. Note that the same domain also serves older plan ladders (a $19 Pro and $199 Team with a per-day credit model) from earlier product generations — if you see different numbers on an old link or a third-party listing, that is why.

My testing notes

Four observations from building with the platform.

First, the describe-to-build flow is the fastest agent onboarding I have used in the no-code category. Writing "qualify inbound leads from the form, enrich with company data, and draft a follow-up" produced a working multi-step agent with sensible tool suggestions; the refinement work was in edge cases, not structure. That is the right trade for the target buyer: an ops manager gets value in an afternoon, and the visual editor is there when the description's defaults are wrong.

Second, the Action metering changes how you think about agent design. Because every tool run counts against the plan allowance, my habits shifted toward agents that batch and reason before acting — one agent run doing five careful tool calls beats five trigger-happy runs. Watching the execution logs helped here: I could see exactly which steps burned Actions and prune wasteful loops. The seven-day task history on the free plan is the sting in the tail: when debugging a misbehaving agent a week later, the evidence is gone. Pro and above are where serious evaluation belongs.

Third, knowledge handling is competent but shallow on the cheap tiers. Ten megabytes of knowledge on free and 100 MB on Pro is enough for product FAQs and a playbook, not a document corpus; the Team and Business tiers raise this, and heavier RAG-style workloads will bump against storage limits before they bump against agent limits. Retrieval quality in my runs was good for well-structured source material and mediocre for messy internal wikis — a pre-processing problem as much as a platform problem.

Fourth, the multi-agent workforce layer is where the product justifies its pricing. Coordinating a researcher agent, a drafter, and a reviewer against one process — with scheduling and an approval habit built into the flow — produced more dependable output than a single agent doing everything, and the analytics dashboard made the improvement visible. This is the part I would evaluate hardest on a trial.

Where it falls short

The no-code ceiling is real: complex branching logic, custom data transformations, and precise retry semantics are harder to express here than in code-based frameworks, and when an agent misbehaves the fix is usually prompt surgery rather than deterministic logic. The dual billing model — Actions plus passed-through vendor credits — means the plan price is not your monthly cost, and high-volume use scales costs in two directions at once. Ecosystem maturity is behind the giants: the integration count is large but premium connectors are paywalled, the public API story is thinner than developer-first platforms, and task history retention on entry tiers is short. The conflicting legacy pricing pages on the vendor's own domain are a trust tax. And as with every agent platform, output quality is hostage to the underlying models — the platform orchestrates, it does not make the reasoning better.

One more thought on the competitive frame: where Zapier gives you deterministic workflow steps with an AI layer bolted on, and Flowise gives you visual control aimed at builders, Relevance AI is betting that the unit of automation is the worker rather than the flow. My experience says that bet pays off as the process gets messier and the judgment calls get more frequent.

Verdict

Relevance AI is the most approachable agent-workforce platform I have tested for non-technical teams: describe the job, wire the knowledge, watch the execution logs, and scale into multi-agent processes when one agent is not enough. I rate it 3.8: core functionality depth is good for business agent work and usability is genuinely strong, while value is qualified by Actions-plus-vendor-credit metering, short free-tier history, and paywalled connectors. For ops, sales, and support teams automating recurring work without engineers, it earns a pilot; for complex logic or high-volume deterministic pipelines, it will fight you.

Sources

• https://relevanceai.com/pricing-new
• https://relevanceai.com/docs
• https://relevanceai.com/workforce
• https://relevanceai.com/careers
• https://relevanceai.com/docs/get-started/pricing
• https://forbes.com.au/news/entrepreneurs/australian-ai-agent-crew-relevance-ai-bag-24-million
• https://forbes.com.au/news/innovation/ai-startup-relevance-15-million-gen-ai
• https://smartcompany.com.au/startupsmart/relevance-ai-raises-15-million-low-code-workplace-platform
• https://inaitoday.com/relevance-ai-raises-au15m-to-help-businesses-scale-rapidly-with-autonomous-ai-workforces
• https://itnews.com.au/news/relevance-ai-secures-15m-in-series-a-for-global-growth-604055
• https://everydev.ai/developers/relevance-ai
• https://deciderstack.com/relevance-ai-review
• https://costbench.com/software/ai-agent-platforms/relevance-ai