Relevance AI

What it does

Writes code

PartlyNot a coding agent, but the built-in Website Builder agent in Chat generates a complete website from a text description; there is no general source-authoring product.Website Builder is a built-in Agent in Chat that generates complete websites from a text prompt or description.relevanceai.com/docs/get-started/chat/c… ↗

Done-for-you service

PartlyYou buy software, but the vendor also puts an embedded deployment team on the first build and then hands the work back to your own team.An embedded deployment team gets you live, then trains your team to build without us.relevanceai.com ↗

How much it does unattended

Runs autonomously

YesAgent-to-tool and agent-to-agent connections can be set to Auto Run so actions execute with no human in the loop, with an optional cap on automatic executions.In this mode, the agent will execute actions automatically without requiring approval.relevanceai.com/docs/build/workforces/w… ↗

Multi-agent

YesWorkforce is a visual canvas for wiring several specialist agents together, and Workforces are unlimited on every published plan tier.Workforce transforms how your agents collaborate by providing a visual canvas where you can design, connect, and monitor teams of specialised agents.relevanceai.com/docs/llms.txt ↗

Agent permissions

YesEach connection carries one of three approval behaviours - Auto Run, Approval Required, or Let Agent Decide - so the autonomy boundary is set per tool, not per product.This mode forces the agent to seek explicit approval before executing an action.relevanceai.com/docs/build/workforces/w… ↗

Which models it runs on

Claude

YesAnthropic Claude models run through Anthropic's API with no data logging, and agent settings expose Claude-specific thinking type and effort controls.Anthropic models are offered through their API, and once again, no data is stored or trained on during this process.relevanceai.com/docs/admin/security ↗

GPT

YesThe vendor's own model/vendor table lists GPT supplied by OpenAI with no logging and no training, available when you supply no API key of your own.| GPT | OpenAI | No | No | Yes |relevanceai.com/docs/admin/security ↗

Open models

YesOpen-weight models are reachable through the OpenRouter integration, and the vendor says it also hosts open-source models such as LLaMA in its own cloud accounts.OpenRouter models are a unified access point to a wide variety of open models like Meta's LLaMA family, Google Gemini models, and others.relevanceai.com/docs/integrations/llm-i… ↗

Model choice

YesYou pick the model per agent, set a fallback model from another provider, and the feature table marks "Bring your own LLM" available on every published tier.relevanceai.com/docs/get-started/pricing ↗

Where you use it

In your editor

PartlyThere is no in-editor coding assistant; instead the MCP server can be added to Cursor, VS Code, Windsurf or Zed to build and run your Relevance agents from the editor.The Relevance AI MCP server gives any MCP-compatible AI client direct access to your agents, tools, and knowledge.relevanceai.com/docs/integrations/mcp/m… ↗

On the command line

PartlyNo first-party CLI of its own, but a published Claude Code plugin bundles the MCP server and skills so agents can be built and managed from a terminal.Use the Relevance AI Claude Code plugin to build and manage your agents directly from your terminal.relevanceai.com/docs/integrations/mcp/c… ↗

In a browser

YesThe product is a hosted web application, with a separate Chat interface documented for desktop, mobile and browser use.Use AI Agents and Workforces in a chat interface on desktop, mobile, and browserrelevanceai.com/docs/llms.txt ↗

Not established: In your pipeline

Whose machine it runs on

Self-hosted

NoHosted multi-tenant service only; customer data is logically separated within the vendor's cloud and a single-tenant option is described as still unbuilt.Relevance operates on a multi-tenanted architecture where customer data is logically separated ... Single-tenant options are currently in the worksrelevanceai.com/docs/admin/security ↗

Not established: Open source

What it costs to run

How it meters

YesTwo meters plus seats: an Action is one tool run, Vendor Credits are the pass-through model and tool cost, and each plan caps how many build users it carries.An Action is a single run of a Tool. Each time a tool runs, it counts as an action — whether it's a simple task like sending one email or running a…relevanceai.com/docs/admin/subscription… ↗

Free tier

NoThe standing $0 plan has been withdrawn: existing Free organisations keep access but no new signup can choose it, so the cheapest way in is the paid Pro tier.The Free plan is retired and closed to new signups. Organizations already on Free keep their existing access and can upgrade at any time.relevanceai.com/docs/get-started/pricing ↗

API access

YesA documented SDK with sk- project API keys for server code plus scoped embed keys for browsers, and agents can additionally be triggered by direct HTTP requests.API keys are available in the project settings on the dashboard. They follow the `sk-...` format and grant unrestricted access to every resource in…relevanceai.com/docs/sdk/authentication ↗

MCP server

YesA hosted MCP server at mcp.relevanceai.com lets another agent list and run your agents, with OAuth access bounded server-side by the caller's project role.The Relevance AI MCP server gives any MCP-compatible AI client direct access to your agents, tools, and knowledge.relevanceai.com/docs/integrations/mcp/m… ↗

Bring your own key

YesBringing your own model keys is optional on every plan and bypasses Vendor Credit metering entirely, with the vendor's credits left as a fallback.To bypass Vendor Credits entirely, you can bring your own API keys. BYOK is optional on every plan — it lets you skip Vendor Credits while still…relevanceai.com/docs/admin/subscription… ↗

Buying it for a team

A company can buy it

YesPaid subscriptions attach to the Organization rather than a person, and Enterprise buyers can request further organisations under one contract.Yes, for Pro and Team plans. Subscriptions are applied at the Organization level, not to individual users or projects.relevanceai.com/docs/get-started/pricing ↗

Seat model

YesSeats are split into build users and end users with hard ceilings per tier - two builders on Pro, five builders plus 45 end users on Team, unlimited only on Enterprise.| **[Build Users](/docs/admin/project-management/add-members)** | 2 | 5 | Unlimited |relevanceai.com/docs/get-started/pricing ↗

Pooled budget

YesOne balance for the whole organisation: everybody shares the plan's Actions and Vendor Credits rather than holding a personal allowance.Everyone in the Organization shares the plan's Actions, Vendor Credits, and feature access.relevanceai.com/docs/get-started/pricing ↗

Admin controls

PartlyPer-connection approval gates and spend/usage limits are available broadly, but role-based and fine-grained access control are gated to Enterprise on the vendor's own tier table.| **[RBAC](/docs/enterprise/rbac)** | ✗ | ✗ | ✓ |relevanceai.com/docs/get-started/pricing ↗

Audit log

PartlyAudit logging exists but only on Enterprise, where events and execution traces are streamed out over OpenTelemetry; Pro and Team are marked as not having it.| **[Audit Logs](/docs/enterprise/streaming-events)** | ✗ | ✗ | ✓ |relevanceai.com/docs/get-started/pricing ↗

Single sign-on

PartlySAML 2.0 and OIDC single sign-on is documented but reserved for Enterprise; the two self-serve tiers are marked without it on the vendor's comparison table.| **[SSO (SAML)](/docs/enterprise/sso-setup)** | ✗ | ✗ | ✓ |relevanceai.com/docs/get-started/pricing ↗

What happens to your code

Opt out of training

YesNothing to opt out of by default: the vendor states it does not train on customer data absent a specific partnership agreement, and metadata is excluded from training.We do not use your data to train our models or improve our services unless you have a specific partnership agreement with usrelevanceai.com/docs/admin/security ↗

Data residency

YesThree regions are offered and chosen at signup - N. Virginia, London or Sydney - but the vendor warns the choice is fixed once the organisation exists.Data is stored in Australia, the US, or the EU/UK, based on your selection at signup.relevanceai.com/docs/admin/security ↗

Getting out

YesData exports in CSV, Excel or JSON are available at any time, and unused Vendor Credits roll over indefinitely while the subscription stays active rather than expiring.You can export your data at any time in standard formats (CSV, Excel, JSON)relevanceai.com/docs/admin/security ↗

Certifications

YesSOC 2 Type II and GDPR are claimed on every tier with third-party assessment reports offered to Enterprise under NDA; no ISO 27001 certification is claimed anywhere read.Relevance AI is SOC2 Type II compliant, and we operate in a multi-region environment.relevanceai.com/docs/admin/security ↗