Relevance AI

The current homepage positions Relevance AI as a platform for specialist agents, multi-agent teams, long-running jobs, triggers, context, tools, evaluations, tracing and an MCP gateway. It presents enterprise controls including approvals, versioning, monitoring, cost visibility, RBAC, SSO, audit logs and PII masking.

The documentation describes low/no-code agents, chat, workforces, knowledge, tools and a marketplace. Agents can send emails, update CRMs, search the web and call APIs; workforces connect agents for multi-step work with triggers, guardrails, approvals and escalations.

Agents plan and complete tasks, select tools, adapt, store successful patterns, communicate with people and agents, and may operate end to end or as copilots. They can escalate unresolved work to a human by email or chat and retain the answer for future reference.

Tools are no-code step chains that call APIs, run LLM prompts, send emails, search databases or execute Python. Execution logs expose inputs and outputs for each step.

Workforce is a visual canvas for connecting and monitoring specialized agents, tools, conditions and triggers as multi-agent systems.

Every agent-to-agent or agent-to-tool edge can auto-run, require approval or let the agent decide based on context and confidence. The task view exposes proposed action and reasoning for approve, reject or guidance; escalations can route context to designated people or agents by Slack or email.

Agent memory extracts metadata from task output and recalls it in later tasks. The broader knowledge system supplies RAG-backed organizational context; these are product-managed memory mechanisms rather than a claim of unconstrained self-learning.

Evals support scenario-based tests and reusable checks for agent behavior and output quality. The pricing comparison reserves agent evaluations for Enterprise.

The Tasks page monitors, troubleshoots and resolves agent and workforce tasks in bulk; the queue exposes task status and cancellation.

Enterprise analytics monitors workforce performance, costs and usage in real time. This is agent-operating analytics, not merchant sales or logistics analytics by default.

Relevance AI documents SOC 2 Type II, GDPR, SSO/MFA, enterprise RBAC/FGA, human-in-loop controls, encrypted keys, configurable retention and multi-region storage in US Northern Virginia, London and Sydney. It says customer data is not used to train models. India is not a listed data-residency region.

Enterprise customers can stream audit logs and execution traces to their infrastructure using OpenTelemetry for compliance, analytics and observability.

Customers can call arbitrary REST endpoints with encrypted project-scoped keys and use official Python or JavaScript SDK access to trigger agents, run tools or manage resources programmatically.

Agents can connect to remote MCP servers, discover and invoke tools and read resources. Presets include Notion, Canva and Atlassian; custom remote Streamable HTTP servers are supported, while local MCP servers are explicitly unsupported.

Relevance AI exposes an MCP server for compatible clients to build and manage Relevance agents and assets; this is developer control, separate from the agent-side MCP client.

Agents can start from scheduled, API, webhook, tool or integration triggers. The documentation describes more than 1,000 integration triggers; plan-specific premium and enterprise trigger limits apply.

The platform connects agents to external services for automated workflows. Current documentation and pricing market more than 2,000 app integrations, while exact scopes and plan restrictions vary by connector and trigger.

A current WhatsApp integration lets agents communicate directly with customers; separate WhatsApp Business documentation covers account interaction. This proves a messaging channel, not catalog, cart, checkout or payment ownership.

The marketplace markets Shopify-oriented agents for customer service, pricing analysis, inventory forecasting, product descriptions, segmentation, recommendations, upselling and fraud analysis. It does not publish connector operations or a live Shopify app listing; its unnamed percentage examples have no customer, cohort or method and are excluded from case proof.

The programmable phone agent speaks with customers and can qualify, collect or verify details, schedule, confirm, route and follow up. Customers control script, tone, data, routing and allowed actions; no ecommerce order, COD/NDR or delivery workflow is specified.

The article discusses retail AI for recommendations, marketing, pricing, inventory, returns, fraud and customer service, but is an educational category guide rather than Relevance product documentation. It is not used to infer direct merchant-system writes.

The current hub surfaces Autodesk, Canva, KPMG, Lightspeed Commerce, Qualified, SafetyCulture, Send Payments and Zembl. The sitemap additionally retains Rakuten Advertising and two blog cases.

Qualified reports 10x output, $7M pipeline and $500,000 closed revenue over six months, 35+ agents, 40-50 BDR tasks reproduced at human-level quality in one week, 60 vendors evaluated and three meetings booked in one afternoon.

The expanded duplicate attributes research, personalized outreach, meeting booking, CRM cleanup and enrichment to the agents. It says one manager now oversees about 25 agents and the team built 150 tools, while 35+ agents operate organization-wide.

Qualified hosts its SVP Revenue Operations with Relevance AI's co-founder. The customer executive says Relevance exceeded 100% stage-one and stage-two attainment in Q4, adds about $17M-$20M in pipeline capacity, and forecasts about 90% stage-two attainment in the first program quarter. The vendor speaker also gives three anonymous deployments with quantified outcomes.

Send Payments reports 40 hours saved weekly, 24/7 operations and thousands of automated conversations through linked customer-response, compliance and CRM agents.

The expanded case gives the 40-hour formula as about eight minutes of CRM notes across 300+ weekly conversations, documents a 12+ hour prior response delay, 100% conversation review, ten unique replies for ten leads, and nontechnical builders creating agents within weeks.

Zembl reports 30% higher customer conversion without BAU headcount growth, 24/7 sales coverage and 60% faster average call time. The case gives 22-minute to roughly 8-9-minute endpoints, almost 2x bill-upload rate within six weeks, a 5/10/15-lead pilot ramp and about 10-11 active agents.

SafetyCulture reports 3x meetings booked, 83% lower cost per meeting and 200% more qualified opportunities after onboarding an AI agent. The page gives 75,000+ customers across 180+ countries as customer context.

Autodesk describes disciplined experimentation, conversion-focused success metrics, reusable knowledge and data-gated scaling, but publishes no quantified outcome or enough detail to isolate Relevance AI's product contribution.

Canva describes agents embedded in Slack and Salesforce for preparation, summarization, enrichment, scoring and repetitive outreach. Coaching, high-trust conversations, negotiation and post-purchase value remain human-led; workflows move from sandbox through human review before governed production. No quantified outcome is disclosed.

The KPMG page is an operating-model interview about standards, ownership, ROI, adoption and production governance. It does not disclose a specific Relevance deployment or quantified KPMG outcome.

Lightspeed's AI BDR researches prospects, drafts outreach, handles replies and hands to reps. Its rollout starts with every output approved and only automates after stable quality; no quantified result is disclosed.

Rakuten Advertising discusses end-to-end and semi-automated agents, human checkpoints, measurable operating models and hub-and-spoke adoption. Its '10x improves odds' and 'one to five' applications are expert guidance, not measured customer outcomes, and are excluded from case metrics.

A guest post by Verisoul's CEO says a two-person growth team operated like 20 within weeks, sent hundreds of custom emails weekly, used Google search in more than half its tools, and reduced a daily research workflow to under five minutes while producing 5-10 leads after screening 30 articles to five.

ActionVFX used bulk tools to generate H1 titles, meta descriptions, product FAQs and expanded catalog descriptions with a fact-checking safeguard. It reports 4,614 fixed SEO issues and an on-page score increase from 40 to 72; publication into the commerce CMS is not described.

The current canonical page exposes only Enterprise with custom Actions and Vendor Credits, unlimited agents/tools/users/projects/workforces, 2,000+ integrations, calling, evaluations, analytics, SSO, RBAC, audit logs and a dedicated account manager. No public enterprise ticket is shown.

The billing model separates Actions from pass-through Vendor Credits, permits bring-your-own model keys and indefinite rollover while subscribed. It says Business was sunset and legacy paid plans were grandfathered into the new model.

A live noncanonical page lists Free at $0 for 200 actions/month, Pro at $29 monthly or $19 annual-equivalent with 2,500 monthly actions, Team at $349 monthly or $234 annual-equivalent with 7,000 monthly actions, and $40 per 1,000 action top-ups. Because the canonical 2026 page is enterprise-only, current general availability of these prices is unresolved.

The migration FAQ says new pricing applied to signups after 8 September 2025 and renewals from 1 December 2025, Business moved to Team, and Salesforce, Snowflake and Zendesk triggers require Enterprise for new connections.

The terms identify OnSearch Pty Ltd, Australian Business Number 33 637 909 409, trading as Relevance AI. Subscriptions are tiered; third-party integrations and customer-provided API keys remain subject to third-party function, cost and terms.

The privacy notice identifies OnSearch Pty Ltd and affiliates/subsidiaries as Relevance AI and covers its website, accounts, products and services.

The company says it has more than 80 employees and names Bessemer, Insight Partners, King River, Galileo and Archangel as investors. The page does not publish revenue, ARR, profitability, valuation or ownership percentages.

Relevance AI announced a $24M Series B led by Bessemer with King River, Insight and Peak XV participating. It names founders Jacky Koh, Daniel Palmer and Daniel Vassilev, reports 40,000 agents created in January 2025, more than 80 staff across Sydney and San Francisco, and identifies Vassilev as co-founder/co-CEO.

Lead investor Bessemer confirms its $24M Series B, names co-CEOs Daniel Vassilev and Jacky Koh plus co-founder Daniel Palmer, and describes Workforce and Invent as current products.

Peak XV says Relevance AI was founded in Sydney in March 2020 by Jacky Koh and Daniel Vassilev, and lists Archangel, Galileo, Insight and King River as co-investors. It omits Daniel Palmer, creating a founder-lineage discrepancy with the 2025 company and Bessemer accounts.

The company launch article describes Invent, Workforce, improved knowledge and more than 2,000 integrations, and identifies Jacky Koh as co-founder/co-CEO. It anchors these modules to the $24M Series B period.

The sitemap contains the nine distinct current customer-story URLs, duplicate expanded Qualified and Send Payments URLs, and the Verisoul and ActionVFX blog cases. No additional canonical customer-case URL was identified in the bounded current sitemap.

The self-serve guide says users can build and test a first agent in under five minutes. This is initial prototyping time, not production deployment time.

Enterprise onboarding covers SSO, directory sync, RBAC, usage controls, event streaming, integration governance, API keys, data retention, per-user integration authentication and limited-beta visual data masking. It recommends directory sync for teams above 20 users.

This company educational page summarizes Qualified as more than $10M pipeline and 50+ agents and SafetyCulture as 50% lower cost per meeting and 2x qualified opportunities. Those values conflict with the canonical case pages' $7M/35+ and 83%/200% wording and are retained as secondary company-page contradictions, not replacements.