AI Agents & Assistants
Relevance AI
Rest of world Report an errorPanel rating · 6 judges · How to read the stars
Category median
Sovereignty: 1 of 4 dimensions proven
0–5 in half steps. 5 means the rubric's top anchor is met on the evidence.
by OnSearch Pty Ltd · relevanceai.com
Report an error on this page Is this your product? →
Link to this page
Add ?theme=dark to the image address for dark backgrounds.
The badge shows the category and the month of the latest evaluation, and EU-Made where
that is the product’s provenance. It never shows a score.
Relevance AI at a glance
- Vendor
- OnSearch Pty Ltd
- Headquarters
- Not stated on the vendor’s pages as of 26 Oct 2026
- Data hosting
- Not stated on the vendor’s pages as of 26 Oct 2026
- Free plan
- Not stated on the vendor’s pages as of 26 Oct 2026
- Panel verdict
- The AI panel rates Relevance AI 3/5, the mean of 6 judges. Relevance AI, a no-code platform for building AI agent teams from the Australian company OnSearch Pty Ltd, is judged strongest on agent capabilities, which scores 8: evaluations gate every build path, human-in-the-loop approvals sit on the visual canvas, and the vendor states its own limit that vector search "always returns results, even if none are truly relevant". how scores are made
- As of
- Vendor pages captured 26 Oct 2026; evaluated by the panel 28 Oct 2026. all sources
Read this page as one judge. Each weighs the same scores by what they care about.
The panel's verdict
Relevance AI, a no-code platform for building AI agent teams from the Australian company OnSearch Pty Ltd, is judged strongest on agent capabilities, which scores 8: evaluations gate every build path, human-in-the-loop approvals sit on the visual canvas, and the vendor states its own limit that vector search "always returns results, even if none are truly relevant". Admin and access scores 7 and integrations and knowledge 6-7, on single sign-on, role-based access, audit exports to the customer's own S3 bucket, and read-and-write SharePoint and Confluence connectors. The split is on processing and models, spanning 4 to 7, the rest at 6: the skeptic scored 4 because no page pins inference regions or EU-only processing, while the automation lead's 7 credits named model families, routing to any provider, and OpenAI, LLaMA and Fireworks hosted in the vendor's own AWS and Azure. The weakest marks fall on sovereignty at 4-5 — the Australian entity under New South Wales law, prompts to the US providers OpenAI and Anthropic, EU/UK storage selectable at signup — and pricing transparency at 2-3, with no price figures on the captured pages.
Speaks for it
- Agent capabilities scores 8, with evaluations gating deployment, A/B testing, live-run sampling and drift flagging documented.
- SharePoint and Confluence connectors document read-and-write actions and event triggers, alongside MCP, custom connectors, an API and workspace-level shared connections.
- Single sign-on via SAML 2.0 and OIDC, identity-provider group sync, role-based access at organization, project and asset level, and audit events exportable to the customer's own S3 bucket are documented.
- Model families are named — Gemini, GPT, Claude, Kimi and GLM — with routing to any provider, bring-your-own API keys, and OpenAI, LLaMA and Fireworks hosted in the vendor's own AWS and Azure environments.
- The security documentation states no training on customer data without a partnership agreement, a vendor table shows OpenAI and Anthropic with no logging and a DPA, and SOC 2 Type II is claimed.
Held against it
- Sovereignty scores 4-5, with the contracting entity OnSearch Pty Ltd under New South Wales law and prompts reaching the US model providers OpenAI and Anthropic.
- The captured pages give different figures for the integration count — 1,000+, 1,800+ and 2,000+.
- The captured pages give different positions on improvement use: the terms grant a licence to use Client Data to improve the Solution, while the security pages state no training or improvement without a partnership agreement.
- Data retention configuration is Enterprise-only, must be enabled by the vendor's team, and does not reach knowledge bases, integrations or API keys.
- Single sign-on is Enterprise-only, one identity provider per organization is the limit, and an email change provisions a fresh account whose permissions and owned assets do not transfer.
Best for
- You are deploying agents that must write into connected systems — creating, updating or deleting SharePoint and Confluence items — with human approval steps on the canvas.
- You want releases gated by evidence: pre-deployment scenarios, A/B testing, live-run sampling and drift flagging before company-wide rollout.
- Your IT team needs enterprise identity controls — SAML/OIDC single sign-on, identity-provider group sync, and audit logs exportable to your own S3 bucket.
- You want model flexibility — named families, routing to any provider and bring-your-own API keys.
Avoid if
- You require a European contracting entity — the vendor is OnSearch Pty Ltd under New South Wales law, and prompts reach the US model providers OpenAI and Anthropic.
- You need published price figures before talking to sales — the captured pricing pages show no per-tier numbers and single sign-on sits behind an Enterprise contact.
- Your users change email addresses routinely — email is the account identifier and a change provisions a new account whose permissions and owned assets do not transfer.
- You need configurable data retention below the Enterprise tier — it is Enterprise-only and requires the vendor's team to enable it.
The scores
| Criterion | Panel median |
|---|---|
| Data use, training & retention | 3.5 of 5 |
| Models & processing location | 3 of 5 |
| Admin, identity & access controls | 3.5 of 5 |
| Integrations, connectors & knowledge | 3.5 of 5 |
| Agent capabilities & guardrails | 4 of 5 |
| European sovereignty | 2 of 5 |
Data use, training & retention
Show reasoningHide reasoning
How this is scored
What happens to prompts, uploaded files, connected data and outputs: whether any of it trains a model (the vendor's or a model provider's), how long it is kept and where logs live, whether zero-retention is available, and whether a data processing agreement covers it.
0 — No statement on training or retention; the privacy policy is a website policy that does not cover the product.
3 — A general statement that business data is "not used for training", with no word on retention, on the model providers behind the product or on logs.
5 — Business data excluded from training by contract, retention periods stated for chats and files, and a DPA offered to business customers.
8 — No-training commitments stated for the vendor and every model provider it routes to, retention configurable by the admin (including deletion of chat history), zero-data-retention terms with the model providers named, and abuse-monitoring retention described.
10 — Data use fully accounted for: training, retention, logging and abuse monitoring stated per model provider, customer-controlled retention and deletion with an audit trail, outputs and connected data treated as customer data in the DPA, and an independent audit (SOC 2, ISO 27001 or ISO 42001) covering the product.
The Head of IT
No-training commitments are published for the platform itself and for the named model providers OpenAI and Anthropic, a data processing agreement is offered, log retention is stated per tier, and Enterprise admins can set a retention window in days covering agent conversations, though the vendor's team must enable it. The captured pages give different positions on improving the solution: the terms grant a licence to use client data to improve it, while a documentation page states data is not used to train or improve services without a partnership agreement. We found no public information on abuse-monitoring retention, or on commitments covering providers beyond those two, even though routing to any model provider is advertised. 5 7 9
The Data Protection Officer
Training is excluded in the vendor's own words for the platform and for the named model providers OpenAI and Anthropic, a DPA is offered, and retention is genuinely configurable — but only as an Enterprise feature the vendor must enable, and it does not reach knowledge bases or data held by subprocessors. The terms grant an irrevocable licence to use Client Data to improve the Solution while the security pages state no training or improvement use without a partnership agreement, so the captured pages give different signals on improvement use. We found no public information on abuse-monitoring retention. 5 6 7 9
The Automation Lead
They state no training on customer data for their own models and publish a vendor table naming OpenAI and Anthropic with no training, no logging and a DPA; run logs have stated retention (30 days on free tier, until you delete otherwise) and an Enterprise data-retention control covers tool runs, agent conversations and versions in configurable day periods. It stops short of the higher bar because retention is Enterprise-only and must be enabled by their team, routing to any provider leaves providers beyond the two named unaccounted for, and the terms separately licence use of Client Data to improve the Solution. 5 6 7 9 13
The Customer-Service Director
The security documentation states our data is not used to train models unless a specific partnership agreement exists, and pairs that with a vendor table showing OpenAI and Anthropic each with no training, no logging and a data processing agreement. Retention is unusually well documented for an agent platform — run logs at 30 days on the free tier, stored until we delete on other tiers, an Enterprise-configurable retention policy in days covering agent conversations, a 60-day account deletion commitment and SOC 2 Type II. The general terms grant a licence to use Client Data "to provide the Solution and to improve the Solution", so the captured pages give different statements on data use for improvement; we found no public information on abuse-monitoring retention or equivalent no-training commitments for the other model families the homepage names. 9 7 5 13
The Works Council Member
A written no-training commitment covers the vendor and a table names OpenAI and Anthropic as neither logging nor training on data, log retention is spelled out (30 days on the free tier, until deleted on other tiers, configurable periods on Enterprise), a data processing agreement is published and a SOC 2 Type II report covers the product. For a works council two things weigh against a higher mark: the Enterprise retention feature must be enabled by the vendor's own team and we found no public information on abuse-monitoring retention or on pseudonymising prompt logs. The terms of service separately grant a worldwide, irrevocable licence to use client data 'to improve the Solution', so the captured pages give different signals on improvement use. 5 6 7 9
The Skeptic
The security documentation states "We do not use your data to train our models or improve our services unless you have a specific partnership agreement with us", a provider table shows OpenAI and Anthropic with no data logging and a data processing agreement, and SOC 2 Type II is claimed. Retention is genuinely documented — run logs 30 days on the free tier and until deleted on paid tiers, plus an Enterprise-only configurable policy covering tool runs and agent conversations but skipping knowledge bases, integrations and API keys — yet we found no public information on abuse-monitoring retention, and the terms separately grant a licence to use Client Data "to improve the Solution", which a buyer should reconcile with the no-training statement before signing. 9 7 5 6 13
Models & processing location
panel disagrees
Show reasoningHide reasoning
How this is scored
Which models sit behind the product and who runs them, where inference and storage happen, whether the customer can pin processing to the EU, choose or exclude model providers, bring its own model or key, or run the product in its own cloud or on premises.
0 — The models behind the product and the processing location are not stated.
3 — The model family is named, but where inference runs and through which provider (the model maker, a hyperscaler, the vendor) is unstated.
5 — Models and their providers named, with an EU processing option stated for storage and at least the default model.
8 — Each available model listed with provider and inference region, EU-only processing selectable by the admin and enforced across models, model providers excludable, and bring-your-own-key or own-cloud deployment documented.
10 — Processing fully under the customer's control and documented: EU-only or on-premises inference for every feature including agents and connectors, open-weight or self-hosted models supported, data flows per feature published, and a residency commitment in the contract rather than on a marketing page.
The Head of IT
Models and their providers are named, with OpenAI and Anthropic as external providers plus OpenAI, LLaMA and Fireworks models the vendor self-hosts in its own AWS and Azure environments, storage is region-locked at signup to the US, EU/UK or Australia, and bring-your-own API keys is supported. The captured pages give different statements on hosting location, with the terms allowing locations outside Australia unless otherwise agreed in writing, and we found no public information on inference regions per model, admin-enforced EU-only processing across models, or single-tenant deployment beyond a statement that it is in the works. 1 5 9
The Data Protection Officer
Models are named down to specific versions, two providers appear in a documented table, customers can bring their own API keys, and the vendor self-hosts OpenAI, LLaMA and Fireworks models within its own AWS and Azure environments. An EU storage region can be selected at signup and is region-locked afterwards, but we found no public information on inference regions per model or on an admin-enforced EU-only setting that covers agents and connectors. The cloud infrastructure provider behind the hosting is unnamed in the captured pages. 1 2 4 5 9
The Automation Lead
Model families and providers are named (Gemini, GPT, Claude, Kimi, GLM; OpenAI and Anthropic in the vendor table), storage region is chosen at signup among the US, EU (London) and Australia, they self-host OpenAI plus open-weight models like LLaMA and Fireworks in their own multi-region AWS and Azure, and routing to any provider with customer-managed keys is supported. The pages name no inference region per model and we found no public information on an EU-only inference setting that is enforced across models; single-tenant is stated to be in the works. 1 5 9 13
The Customer-Service Director
The models are named — Gemini, GPT, Claude, Kimi and GLM families, with a table pairing GPT with OpenAI and Claude with Anthropic — and we can bring our own API keys, route to any provider, and use models Relevance hosts itself on AWS and Azure. Storage is selectable at signup between US (N. Virginia), EU (London) and AU (Sydney) and locked thereafter, which satisfies my EU storage requirement. We found no public information on the inference region per model or EU-only enforcement across every feature, the cloud infrastructure provider is unnamed, and single-tenant deployment is only described as in the works. 1 9 2 5 4
The Works Council Member
The model families are named (Gemini, GPT, Claude, Kimi, GLM), routing to any provider is supported, the vendor itself hosts OpenAI, LLaMA and Fireworks models in its own multi-region AWS and Azure environments, and customers can bring their own API keys. A storage region — US, EU/UK or Australia — is selected at sign-up and locked thereafter. We found no public information on the inference region per model or on an EU-only processing guarantee enforced across models, and single-tenant deployment is stated to be in development. 1 5 9 13
The Skeptic
Model families are named (Gemini, GPT, Claude, Kimi and GLM among them), routing to any provider is advertised, and bring-your-own API keys plus vendor-hosted OpenAI, LLaMA and Fireworks in the vendor's own AWS and Azure environments are documented. But nothing states where inference runs for each model — only that data is stored in Australia, the US or the EU/UK based on a choice made at signup — and we found no public information pinning even the default model's inference to the EU. 1 9 2 5
Admin, identity & access controls
Show reasoningHide reasoning
How this is scored
What the IT department controls: SSO and SCIM provisioning, roles and groups, which models, agents and connectors each group may use, retrieval that respects the permissions of the source system, audit logs, usage reporting, and policies such as data-loss prevention — and whether reporting on staff use is scoped and documented.
0 — No admin features documented; accounts are individual sign-ups.
3 — A team workspace with an owner and member invitations, but no SSO, no roles and no audit log.
5 — SSO, admin and member roles, user management, and an admin view of usage.
8 — SAML/OIDC SSO with SCIM provisioning, group-based permissions for models, agents and connectors, retrieval that inherits source-system permissions, exportable audit logs, and usage analytics with documented scope.
10 — Governance a regulated company can operate: fine-grained policies (DLP, content filters, prompt and output logging rules), audit logs streamed to a SIEM, per-agent ownership and approval, usage reporting that can be aggregated or pseudonymised for works-council agreements, and every admin action logged.
The Head of IT
SAML 2.0 and OIDC single sign-on runs through WorkOS with Okta, Microsoft Entra and Google Workspace, identity-provider groups sync for automated provisioning, and permissions operate at organization, project and asset levels with fine-grained access control for Enterprise customers; audit logs export to S3 and stream over OpenTelemetry, which my SIEM can ingest. Single sign-on is Enterprise-only and limited to one identity provider per organization, and an email change provisions a new account rather than transferring permissions or owned assets, a lifecycle wart at my scale. We found no public information on retrieval that inherits source-system permissions such as SharePoint's, or on model entitlement per group. 7 9 10
The Data Protection Officer
SAML and OIDC single sign-on via WorkOS, identity-provider group sync for automated provisioning, role-based control at organization, project and asset level, fine-grained access for Enterprise, and audit logs exportable to the customer's own S3 bucket are all documented, alongside prompt-injection detection with traces streamed over OpenTelemetry. However, single sign-on and fine-grained access are Enterprise-only. We found no public information on retrieval that inherits source-system permissions or on the documented scope of usage reporting on staff. 7 9 10
The Automation Lead
SAML 2.0 and OIDC SSO via WorkOS with identity-provider group sync and automated provisioning, role-based access at organization, project and asset level plus fine-grained access for Enterprise, audit logs streamable to your own S3 bucket, OpenTelemetry event streaming, and an API for managing projects and users — that is most of what my IT side needs. But SSO is Enterprise-only, we found no public information on retrieval that inherits the permissions of the source system, and email is the identity anchor: an email change provisions a fresh account with no permissions or owned assets transferred. 3 7 9 10
The Customer-Service Director
SSO with SAML 2.0 and OIDC, MFA, directory sync of identity-provider groups for automated provisioning, and role-based access at organisation, project and asset level — plus fine-grained access control for Enterprise — is exactly what my IT team would ask for. Audit events export to our own S3 bucket and stream over OpenTelemetry, with prompt-injection detection, PII masking and visual data masking on top. We found no public information on retrieval that inherits the permissions of the source system, the scope of the usage analytics is not detailed on the captured pages, and SSO is restricted to the Enterprise plan. 10 9 7 3
The Works Council Member
SAML 2.0 and OIDC single sign-on, identity-provider group sync for automated provisioning, role-based access at organization, project and asset level, and exportable audit logs streamed to the customer's own S3 bucket are all documented. Two gaps matter for a co-determination agreement: we found no public information on whether usage reporting on staff can be aggregated or pseudonymised, and no public information on whether retrieval inherits the permissions of the source system. Single sign-on, fine-grained access and prompt-injection detection are Enterprise-only. 7 9 10
The Skeptic
The Enterprise tier has SAML 2.0/OIDC single sign-on via WorkOS, identity-provider group sync for bulk provisioning, and role-based permissions at organization, project and asset level, with audit events exportable to the customer's own S3 bucket and OpenTelemetry streaming, plus Enterprise prompt-injection detection. Two operational warts: single sign-on is Enterprise-only, one identity provider per organization is the limit, and a user whose email changes is provisioned a fresh account whose permissions, group memberships and owned assets do not transfer. We found no public information on retrieval that inherits the source system's permissions. 10 9 3 7
Integrations, connectors & knowledge
Show reasoningHide reasoning
How this is scored
How the product reaches company knowledge and systems: connectors to file stores, wikis, mail, CRM, ticketing and ERP, how fast and how completely they sync, the API, support for open protocols such as MCP, and how the assistant cites the sources it answers from.
0 — No integrations; users paste text or upload files one at a time.
3 — File upload and one or two connectors (for example a drive), with no API and no statement on sync or citations.
5 — A connector library covering the common file stores, wikis and collaboration tools, answers citing sources, and a documented API.
8 — Connectors to file stores, wikis, mail, CRM and ticketing with incremental sync, custom connectors or MCP support, an API and webhooks for embedding assistants and agents in other tools, and source citations on every grounded answer.
10 — The product sits in the company's systems: read and write connectors with documented scopes, MCP servers and clients, contact-centre or telephony integration where relevant, a knowledge layer that can be audited and corrected, and the data, agents and configuration exportable when the customer leaves.
The Head of IT
Confluence and SharePoint connectors ship read and write steps with event triggers, native triggers cover CRM, email and calendar alongside webhooks and cron, custom connectors and MCP are supported, connections are shared once at workspace level, and data exports in CSV, Excel and JSON. The captured pages give different figures for the size of the connector library, quoting 1,000+, 1,800+ and 2,000+. We found no public information on source citations for grounded answers or on sync cadence and completeness for connected systems. 3 9 11 12 13
The Data Protection Officer
A thousand-plus native connectors plus MCP and custom connectors are documented, with read and write actions and triggers spelled out for SharePoint and Confluence, an API with webhooks, and native CRM, email and calendar triggers. Export paths exist in standard formats plus OpenTelemetry and Delta Share export, and the knowledge layer states its own limits such as vector search always returning results. We found no public information on source citations on grounded answers or on sync speed and completeness for the connectors. 1 11 12 13 14
The Automation Lead
MCP support, custom connectors, an API, webhooks, cron and native CRM, email and calendar triggers, plus read-and-write Sharepoint and Confluence actions with event triggers and workspace-level shared connections — this is the connective tissue I build agents on. The captured pages give different figures for the integration count (1,000+, 1,800+ and 2,000+), we found no public information on sync speed or source citations on grounded answers, and the knowledge-search docs themselves say vector search always returns results even when none are truly relevant. 1 11 12 13 14
The Customer-Service Director
The connectors are deep rather than broad-only: the SharePoint and Confluence pages document read and write actions plus list and page triggers, and there are native CRM, email and calendar triggers, webhooks, cron, 1,000+ apps, custom connectors, MCP support and workspace-level connections every agent can share. That is the shape I need for order and ticket systems, though we found no public information on a specific ticketing connector, on incremental sync behaviour, or on source citations for grounded answers — the quality-control material describes query refinement and result validation but not citations showing which document an answer came from. 12 11 13 14 1
The Works Council Member
A large connector library is documented with native triggers, webhooks and cron, MCP support, custom connectors, a public API, and read-and-write actions such as creating, updating and deleting SharePoint and Confluence items from workspace-level shared connections. The captured pages give different figures for the connector count (1,000+, 1,800+ and 2,000+), which we do not choose between. We found no public information on source citations for grounded answers or on sync cadence. 3 11 12 13 14
The Skeptic
Connectors are real and read-write — the SharePoint and Confluence pages list specific create, update and delete actions plus list-item and page triggers, native triggers cover CRMs, email and calendar alongside webhooks and cron, and MCP, custom connectors and an API are offered, with knowledge search using query refinement and vector search over knowledge tables. The captured pages give different figures for the integration count — 1,000+, 1,800+ and 2,000+ appear on different pages. We found no public information on sync completeness, incremental sync, or citations on grounded answers. 13 11 12 14 3
Agent capabilities & guardrails
Show reasoningHide reasoning
How this is scored
What an agent can actually do and where it stops: multi-step tasks, tool use and actions in other systems, scheduling and triggers, human approval before consequential actions, run logs, evaluation and testing before release, and the limits the vendor states rather than hides.
0 — No agents; the product is a chat window, whatever the marketing says.
3 — Saved prompts or custom assistants with instructions and files, but no actions in other systems and no run history.
5 — Agents that call tools or APIs and run multi-step tasks, built without code, with a log of each run.
8 — Agents that act in connected systems with scoped permissions, human approval configurable per action, triggers and schedules, test and evaluation tooling before release, and documented limits and failure handling.
10 — Agents a business can put in production and account for: versioning and staged rollout, evaluation against test sets, monitoring of quality and cost per agent, handover to a human with full context, every action attributable to the agent and the person who deployed it, and the vendor's own statement of what agents must not be trusted with.
The Head of IT
Agents act in connected systems through workspace-level connections with key scopes the customer manages at the underlying integration, human-in-the-loop approvals sit on the visual canvas alongside escalation monitoring, triggers and schedules come via webhooks, cron and native events, and every build path ships gated by the same evals, with live-run sampling, pass-rate charting and drift flagging before release. Version control, A/B testing, real-time monitoring and cost visibility per agent and per task are documented, and the vendor plainly states that vector search always returns results even when none are truly relevant, the kind of stated limit I want to see. We found no public information on staged rollouts or on attributing every action to the agent and the person who deployed it. 1 2 3 9 13 14
The Data Protection Officer
Agents act in connected systems with documented read and write actions, human-in-the-loop approvals are a first-class building block, native triggers, webhooks and cron drive scheduling, and evaluations gate deployment with sampled live runs, pass-rate charting and drift flagging. Run logs are retained and exportable, version control is documented, and cost visibility per agent is stated. We found no public information on attributing every action to the person who deployed the agent or on a vendor statement of what agents must not be trusted with. 1 9 11 12 13 14
The Automation Lead
Agents act in connected systems with human-in-the-loop approvals you drag into the flow, triggers and schedules, version control with protected draft and active versions, full agent tracing with cost visibility, agent evaluations, A/B testing, pre-deployment scenarios and live-run sampling with pass-rate charting and drift flags — the build-test-release loop I demand before company-wide rollout. We found no public information on staged rollout, per-action attribution to the deploying person, or the vendor's own statement of what agents must not be trusted with, and they state tool inputs and outputs are not logged by Relevance AI. 1 2 3 7 9 13 14
The Customer-Service Director
This is the strongest area for a contact centre: agents run multi-step tasks in connected systems with human-in-the-loop approvals on a visual canvas, triggers via webhooks, cron and CRM/email/calendar events, and — what I care about most — evaluation before go-live, with agent evaluations, pre-deployment scenarios, A/B testing, an eval gate held at a 90% bar, and live-run sampling with pass-rate charting and drift flagging. Version control, full agent tracing, cost visibility and S3 audit events give me attributable runs, with escalation and "every task, decision, and escalation visible in one place". The vendor even states a limit of its own — vector search "always returns results, even if none are truly relevant" — though we found no public information on handover to a human carrying the full conversation context stated explicitly. 13 3 2 9 14
The Works Council Member
Human-in-the-loop approvals appear as an explicit step on the build canvas, with oversight of every task, decision and escalation in one view — precisely the control a works council needs before an agent touches a colleague. Every build path is gated by evaluations, with pre-deployment scenarios, live-run sampling, drift flagging, version control, full tracing and cost visibility documented, and limits are stated in places (vector search 'always returns results, even if none are truly relevant'; an irreversible Confluence delete action). We found no public information on attribution of each action to the agent and the person who deployed it. 1 2 9 13 14
The Skeptic
Agents run multi-step on a no-code visual canvas, approvals can be dragged into the flow, every run is traced, cost per agent is visible, and evaluation is treated as first-class: agent evaluations, A/B testing, sampling of live runs with drift flagging, and the vendor's own claim of a 96.4% eval pass rate held against a self-set 90% bar, with versioning and human-in-the-loop escalation documented. Credit where due on stated limits — the academy page says plainly that vector search "always returns results, even if none are truly relevant" — though we found no broader statement of what agents must not be trusted with, and no per-action attribution to the deployer. 1 13 2 3 14
European sovereignty
panel opinion
Show reasoningHide reasoning
How this is scored
Who the contracting entity and the parent company are, where customer data is hosted, and who the subprocessors are — model providers above all, then hosting, search, telephony and email. Independently sourced by the sovereignty pipeline; for an AI product the model provider is a subprocessor, so a European interface over a non-European model counts as such.
0 — Non-EU vendor and contracting entity, hosting unstated, subprocessors (including model providers) unnamed.
3 — A privacy policy exists, but hosting location is unstated or non-EU and the model providers behind the product are not named as subprocessors.
5 — EU hosting stated and an EU contracting entity, but prompts reach non-EU model providers or other subprocessors without a stated safeguard or EU region.
8 — EU hosting on named infrastructure, EU contracting entity, every subprocessor listed with its location including model providers, EU inference available, and a DPA offered.
10 — Sovereign end to end and evidenced: European ownership, entity, hosting and model processing, every subprocessor European or optional, data export and deletion documented, and a security or certification statement published rather than implied.
The Head of IT
The contracting entity is OnSearch Pty Ltd trading as Relevance AI under New South Wales law with exclusive jurisdiction there, and the model providers processing prompts are OpenAI and Anthropic. Customers select US, EU/UK or Australia storage at signup, subprocessors are named by region across the UK, EEA, US and Australia with a public list, standard contractual clauses are referenced and a data processing agreement is offered. The contracting entity is Australian rather than European, and self-hosted inference runs in the vendor's own AWS and Azure environments rather than a European sovereign option. 4 5 6 9
The Data Protection Officer
The contracting entity is OnSearch Pty Ltd in Australia under New South Wales law, hosting can be pinned to an EU region at signup, and model providers OpenAI and Anthropic are named with a DPA — but the model providers and other subprocessors sit across the US, UK, EEA and Australia, and a public subprocessor list exists whose contents are not among the captured pages. The cloud infrastructure provider is unnamed, and we found no public information on EU-only inference for models, agents or connectors. Standard Contractual Clauses and SOC 2 Type II are referenced. 4 5 6 9
The Automation Lead
The contracting entity is OnSearch Pty Ltd in Australia under New South Wales law, hosting is selectable at signup among the US, EU (London) and Australia, a subprocessor list is published at their trust centre, and the privacy policy names sub-processors in the UK, EEA, US and Australia including US model providers OpenAI and Anthropic with a DPA. No European entity or ownership is documented, the region is locked once at signup, and prompts can reach non-EU providers, which keeps it in the lower band. 5 6 9
The Customer-Service Director
The contracting entity is OnSearch Pty Ltd trading as Relevance AI, an Australian company under New South Wales law, with the cloud infrastructure provider unnamed in the security policy. Customer data storage is selectable between Australia, the US and the EU/UK at signup, sub-processors are stated to sit in the UK, EEA, US and Australia, a subprocessor list URL is published, standard contractual clauses are cited and a DPA is offered — but the named model providers are the US companies OpenAI and Anthropic, and the content of the subprocessor list is not among the captured pages. This is a non-European vendor with an EU storage option, not a sovereign European offering. 5 6 9 13
The Works Council Member
The contracting entity is OnSearch Pty Ltd, trading as Relevance AI, in Australia under the exclusive jurisdiction of New South Wales courts; hosting sits with an unnamed cloud infrastructure provider, an EU/UK storage region is selectable at sign-up, and the model providers named include OpenAI and Anthropic in the United States. Standard Contractual Clauses are cited for transfers and a data processing agreement is offered, and a subprocessor list URL is published, but the captured pages do not show the list itself. We found no public information on EU inference or on the vendor's ownership structure. 4 5 6 9
The Skeptic
The contracting entity is OnSearch Pty Ltd, an Australian company under New South Wales law whose terms allow hosting "potentially other locations outside Australia", and prompts reach US model providers OpenAI and Anthropic — named in the admin documentation with data processing agreements — while the privacy policy relies on Standard Contractual Clauses for transfers and discloses sub-processors in the UK, EEA, US and Australia. An EU/UK storage region can be selected at signup, but the cloud infrastructure provider is unnamed, the full subprocessor list was not among the captured pages, and we found no public information on European ownership or EU inference. 5 9 6
Pricing transparency
not rated — the vendor publishes no price
Show reasoningHide reasoning
How this is scored
Whether a company can compute the real annual cost from public pages alone: price per seat, per agent, per conversation or per usage unit, what is included in each tier, usage limits and overage, which models, connectors or agents cost extra, minimum seats and term.
0 — No public prices at all; every customer is a sales call.
3 — A starting price or "from" figure exists, but usage limits, included models and what scales the cost are unclear.
5 — Prices per tier are public with the main inclusions, but usage limits, overage, agent runs or enterprise features (SSO, audit logs, EU processing) are unpriced or "on request".
8 — Every tier priced publicly with seat or usage units, usage limits and overage rates, add-ons listed, minimum seats and term stated.
10 — Complete price computability: annual cost derivable for a given number of seats, agents or conversations, with every usage rate, model surcharge, add-on and support tier and the cancellation and data-return terms published.
The Head of IT
The terms publish the fee structure, being subscription fees, actions where any tool or agent run counts as activity, and vendor credits charged at cost with GST excluded, and a free account and free trial exist, with the pricing page listing tier inclusions such as unlimited users, agents and workforces. No per-seat or per-run figures appear on the captured pages, single sign-on sits behind an Enterprise sales conversation, and I cannot compute an annual cost for eight hundred staff from public pages alone. 3 5 8 10 12
The Data Protection Officer
A self-serve free account is documented, and the fee structure is public in the terms — subscription fees, actions and vendor credits charged at cost with no markup — with the metering unit defined: any execution of a tool or agent counts as activity, manual or scheduled. We found no public information on price figures for the paid tiers, on usage limits, or on the cost of Enterprise features such as single sign-on, data retention or EU hosting. 3 5 8 12
The Automation Lead
A public pricing page exists and the terms define the fee shape — subscription fees, paid actions where any manual, automated or scheduled tool or agent execution counts as activity, and vendor credits passed through at cost — but the captured pages show no price figures per tier and we found no public information on usage limits or overage, while SSO sits behind an Enterprise sales contact. I cannot compute an annual cost from what was captured. 3 5 8 10
The Customer-Service Director
The fee structure is named — subscription fees, actions, and vendor credits charged at cost — and the usage unit is defined: any execution of a tool or agent, manual or scheduled, counts as activity, with unlimited users, agents, tools and workforces advertised and a free account available. But the captured pricing pages show no per-tier prices, no usage allowances and no overage rates, and SSO comes only with an Enterprise plan you must discuss with the sales team. My CFO cannot forecast a cost per conversation from the public pages as captured. 5 3 8 10 12
The Works Council Member
A free account exists and the terms state that fees consist of subscription fees, actions and vendor credits charged at cost, with any execution of a tool or agent counting as activity — so what scales the cost is documented. However, the captured pricing page shows no price figures, and we found no public information on per-tier prices, usage limits, overage rates, minimum seats or contract term. Enterprise capabilities such as single sign-on are routed to the sales team. 3 5 8
The Skeptic
The terms name three cost components — Subscription Fees, Actions, and Vendor Credits charged at cost without markup — and define the billing unit as any manual or scheduled execution of a tool or agent, but no captured page shows a price figure for any tier; the pricing page lists inclusions such as unlimited agents and a dedicated account manager without numbers. Single sign-on and data residency sit behind an Enterprise sales contact. A buyer cannot compute an annual cost from anything we read. 3 5 8 12 10
European sovereignty — proven facts
1 of 4 dimensions provenBuilt only from facts shown on the vendor's own pages. A dimension we could not prove is left open, not scored as zero.
| Legal entity | Not determined | — | uncited Report an error |
|---|---|---|---|
| Ownership | Not determined | — | uncited Report an error |
| Data residency | Not determined ⚠ unverified | — | uncited Report an error |
| Subprocessors | US CLOUD Act reach ⚠ unverified | 0/2 pts | 9 Report an error |
Questions about Relevance AI
Is Relevance AI European?
No. whats-best.ai classifies Relevance AI as Rest of world.
Where does Relevance AI store data?
Not stated on the vendor’s pages as of 26 Oct 2026.
Does Relevance AI have a free plan?
Not stated on the vendor’s pages as of 26 Oct 2026.
What are the best alternatives to Relevance AI?
As of 28 Oct 2026, the panel rates Langdock (4/5), Dust (3.5/5) and Unique AI (3.5/5) highest among the other 15 AI Agents & Assistants products evaluated. Ratings that look equal are rounded to half stars; the order uses the unrounded rating, then the name. The highest-rated European alternative is Langdock (EU-Made). Ratings are the AI panel’s opinion, formed from public evidence: how scores are made. The full list, with European products first: Relevance AI alternatives.
Where this could be wrong
- Evidence ages. The oldest capture behind this page is from 26 Oct 2026. Vendors change pricing and policies without notice; every fact reflects its source as of the capture date shown in the registry.
- Weak sourcing — Data residency. Not confirmed on the vendor’s own pages as captured.
- Weak sourcing — Subprocessors. The full subprocessor list at trust.relevanceai.com/subprocessors is not among the excerpts, and the privacy policy additionally names sub-processors in the UK, EEA, US and Australia.
- AI can misread a source. Extraction and judgement are automated; a citation guarantees traceability, not infallibility. If something here is wrong, say so — no account needed, every report is decided within 5 business days, and accepted corrections are published.
What we left out
A claim that does not survive our checks costs us the claim, not the page. This is what was taken off this one.
- 6 product facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 compliance facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 legal facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 pricing facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 1 integrations fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 support fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 sovereignty dimension could not be confirmed on the vendor’s own pages and is shown as unknown. Know more? Tell us
Sources (14)
The pages every claim on this page was read from — each one checked, dated, and kept verifiable.
- 1 Vendor homepage relevanceai.com Checked 26 Oct 2026 Details →
- 2 Features page (agents) relevanceai.com Checked 26 Oct 2026 Details →
- 3 Pricing page relevanceai.com Checked 26 Oct 2026 Details →
- 4 Data security policy relevanceai.com Checked 26 Oct 2026 Details →
- 5 Terms naming the legal entity relevanceai.com Checked 26 Oct 2026 Details →
- 6 Privacy policy relevanceai.com Checked 26 Oct 2026 Details →
- 7 Data use, training & retention — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 8 Data use, training & retention — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 9 Admin, identity & access controls — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 10 Admin, identity & access controls — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 11 Integrations, connectors & knowledge — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 12 Integrations, connectors & knowledge — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 13 Agent capabilities & guardrails — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →
- 14 Agent capabilities & guardrails — found from sitemap relevanceai.com Checked 26 Oct 2026 Details →