Relevance AI
One of the few agent platforms shipping retrieval, evals and tracing in one box — then it puts SSO, RBAC, audit logs and the evals behind the Enterprise quote, so the €215-a-month plan you can actually buy online is the one without governance.
PRICING
| Free | $0/mo | 200 Actions/month, 1,000 one-time Vendor Credits, 1 build user, 1 project, 1 Workforce, 30-day task history, marketplace access, SOC 2 and GDPR |
| Pro | $19/mo | Annual billing; $29 month-to-month. 2,500 Actions and $20 Vendor Credits a month, 2 build users, unlimited Workforces, scheduled tasks, escalations, premium triggers, bring-your-own LLM, 90-day history |
| Team | $234/mo | Annual billing; $349 month-to-month. 7,000 Actions and $70 Vendor Credits a month, 5 build users and 45 end users, 5 shared projects, calling and meeting agents, A/B testing, analytics dashboard, credit rollover |
| Enterprise | Custom | Custom Actions and credits, unlimited users and projects, Agent Evaluations, work-hour controls, multi-org management, SSO / RBAC / audit logs, enterprise triggers (Salesforce, Snowflake, Zendesk), data retention policies, dedicated account manager |
| Top-ups | $80 / 1,000 Actions · $20 / 10,000 Vendor Credits | Paid plans only; Free users must upgrade first |
Two meters, not one. An Action is counted every time an agent runs a tool, whether that tool sends one email or executes a twenty-step workflow — so cost tracks tool calls, not complexity. Vendor Credits cover the underlying LLM and tool spend at wholesale, and bringing your own API keys on a paid plan removes that meter entirely. Subscriptions apply at organisation level, not per user, and cannot be transferred between organisations.
checked 2026-08-20 · vendor pricing page
Element scores
Strengths
Relevance ships as one system what most teams assemble from six vendors: triggers, a context layer, an MCP gateway, an LLM router, evals, agent orchestration, a job queue and tracing. The evals are the standout — drafted from your own historical cases, re-run on every change, blocking a failing version from publishing, and sampling production runs so quality drift surfaces in days rather than quarters. Retrieval is a real product, not a checkbox: selectable embedding models, configurable chunking, reranking, and automatic syncing from Drive, SharePoint and Notion. The operational plumbing is the part that separates a demo from a deployment — managed queue, concurrency caps, retries with backoff, a dead-letter queue, work-hours pacing. And protocol support runs both ways, including a Claude Code plugin and an agent-skills repo.
Honest dings
The governance story is priced, not built in. RBAC, fine-grained access control, SSO, audit logs, work-hour controls and Agent Evaluations are all Enterprise-only, so the plans you can buy with a credit card are the ones without them — and the product page markets evals as something every agent ships with. Task history is 30 days on Free and 90 on Pro and Team, which is thin if you need to reconstruct what an agent did last quarter. Data residency offers Australia, the US or 'EU (London)' — London is not in the EU, and for a European buyer with GDPR transfer questions that label is doing more work than it should. Region is fixed at signup and cannot be changed afterwards. Single-tenant is 'in the works'. And the two-meter pricing model is the most common complaint from users at production volume: Actions and Vendor Credits deplete at different rates, so forecasting takes real attention.
Sources (17) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- relevanceai.com/docs/get-started/pricing — Official pricing: plan tiers, Actions and Vendor Credits per plan, build and end user counts, task history retention, top-up rates ($80/1,000 Actions, $20/10,000 credits), rollover rules, and the full feature comparison showing which capabilities are Enterprise-only (accessed 2026-08-20)
- relevanceai.com/product — Product page: self-description, the nine feature groups (instructions, models, tools, knowledge and memory, autonomy and approval, triggers, channels, oversight, multi-agent orchestration), eval workflow, managed queue behaviour, enterprise controls list including OTEL and Delta Share export, and named customers (accessed 2026-08-20)
- relevanceai.com/docs/admin/security — Security overview: SOC 2 Type II and GDPR, data residency options (US N. Virginia, EU London, AU Sydney) fixed at signup, 30-day free-tier log retention, encryption, RBAC/FGA/SSO as Enterprise features, prompt-injection detection on OTEL traces, visual data masking, self-hosted OpenAI and open-source models, no training on customer data (accessed 2026-08-20)
- relevanceai.com/docs/get-started/core-concepts/know… — Knowledge as the documented RAG system: semantic search, retrieval flow, add-all-to-prompt versus agent-searches, sources including files, websites, Google Drive, SharePoint and Notion (accessed 2026-08-20)
- relevanceai.com/docs/build/tools/knowledge/knowledge — Embedding specifics: all-mpnet-base-v2 (768 dimensions) as the default for embedding and retrieval, vectorisation at upload, advanced step embeds on the fly and caches (accessed 2026-08-20)
- relevanceai.com/docs/build/tools/tool-steps/knowled… — Advanced Knowledge Search configuration: Cohere embed-english-v3.0 and embed-multilingual-v3.0, chunk sizes 128/256/512/1024, chunking strategy, reranker model, post-processing LLM (accessed 2026-08-20)
- relevanceai.com/docs/integrations/mcp/mcp-server — MCP server: supported clients (Claude Desktop, Claude Web, ChatGPT, Cursor, VS Code, Windsurf, Codex, Zed, v0), per-project authentication, run-only access mode (accessed 2026-08-20)
- relevanceai.com/docs/integrations/mcp/mcp-client — MCP client: agents connect to remote MCP servers over Streamable HTTP, preset connections for Notion, Canva and Atlassian, multiple labelled instances of one server for staging versus production (accessed 2026-08-20)
- relevanceai.com/docs/admin/subscriptions/plans — Plans and credits mechanics: what an Action is, credit-card-only payment, top-up increments, rollover behaviour by plan (accessed 2026-08-20)
- relevanceai.com/docs/llms.txt — Documentation index confirming the surface: MCP server and client pages, Claude Code plugin, agent skills repository, subagents, skills, evals, JavaScript SDK (accessed 2026-08-20)
- bvp.com/news/meet-the-founders-of-relevance-ai — Bessemer's $24M Series B announcement with founders Daniel Vassilev, Jacky Koh and Daniel Palmer; Workforce and Invent launch context (accessed 2026-08-20)
- startupdaily.net/topic/funding/ai-agents-builder-tr… — Independent report of the US$24M (A$37M) Series B, Sydney headquarters, founder quotes on the no-code multi-agent positioning (accessed 2026-08-20)
- capitalbrief.com/briefing/no-code-ai-agent-builder-… — Series B coverage naming Qualified, Activision and SafetyCulture as customers, and describing Workforce and Invent (accessed 2026-08-20)
- coldiq.com/blog/relevance-ai-pricing — Independent pricing breakdown corroborating the annual versus monthly split and the dual Actions / Vendor Credits meter (accessed 2026-08-20)
- checkthat.ai/brands/relevance-ai/pricing — Independent pricing analysis: September 2025 restructure that split one credit pool into two meters, and the recurring user complaint about forecasting at production volume (accessed 2026-08-20)
- cybernews.com/ai-tools/relevance-ai-review — Hands-on independent review of the Workforce canvas, marketplace, agent run modes and the observation that Invent's draft agent needs manual refinement for complex tasks (accessed 2026-08-20)
- tracxn.com/d/companies/relevance-ai/__GcnNqHYzkszu5… — Company profile: founded 2020 in Sydney, $37M total funding across seed (Dec 2021), Series A (Dec 2023) and Series B (May 2025) (accessed 2026-08-20)