The enterprise-grade agent platform

Build, run and manage AI agents at scale. The controls to keep them in production.

Book a demo Try it out

Inbound email

New Salesforce lead

Message in #support

Every weekday, 9am

Webhook received

Update CRM record

Send tailored reply

Resolve ticket

Book meeting

Post summary

Inbound email

New Salesforce lead

Message in #support

Update CRM record

Send tailored reply

Resolve ticket

Building agents is the easy part...How do you put them to real work?

How do you trigger agents at the right time, with the right data and tools?

How do you orchestrate specialist agents to reliably complete tasks?

How do you evaluate, monitor and improve agent quality at enterprise scale?

How do you give agents secure, governed access to your systems?

All of your agents on one stack

Most teams bolt together a router, a queue, an eval tool and a tracer just to run agents. Relevance ships it all as one system.

Triggerslike Zapier
Context layerlike Zep
MCP Gatewaylike Composio
LLM routerlike OpenRouter
Evalslike Braintrust
Agent builderlike Dust.tt
Agent orchestrationlike CrewAI
Job queuelike Temporal
Tracinglike Langfuse

On one platform, the pieces make each other better

Catch regressions before they ship

Agents+Evals

Run on the cheapest model that passes

LLM router+Evals

Score live runs, not just test sets

Tracing+Evals

Govern tool access in one place

MCP Gateway+Agents

Failed runs retry instead of disappearing

Agent orchestration+Job queue

See what every task really cost

LLM router+Tracing

Catch regressions before they ship

Agents+Evals

Run on the cheapest model that passes

LLM router+Evals

Score live runs, not just test sets

Tracing+Evals

Govern tool access in one place

MCP Gateway+Agents

Failed runs retry instead of disappearing

Agent orchestration+Job queue

See what every task really cost

LLM router+Tracing

Build one context layer that all of your agents draw from

Define your tone of voice, business context and knowledge once, not in every agent.

100s of features needed to specialize and optimize agents

We've worked for years with enterprise to deploy agents and we know all of the things you need to be successful.

View all features

Instructions & identity

System prompt (rich markdown)·Prompt description for parent agents·User-facing usage guide·Auto-title prompt

Models

Any major provider + bring-your-own·Fallback model on failure·Temperature & max output tokens·Reasoning & thinking modes·Strict & parallel tool calling·Switch model after N tokens

Tools & actions

Tested integration steps·Sub-agents & MCP servers·Sandboxed Python & JavaScript·Per-tool approval rules·Retry & execution limits

Knowledge & memory

Long-term memory (project and user)·Compaction & observational memory·Hybrid RAG with reranking·Automatic syncing with key data sources·Custom metadata filled by agents

Autonomy & approval

Autonomy step limits·Ask-approval or terminate on limit·Conditional approval rules·Cost-based pauses·Force retry on error

Triggers & scheduling

Recurring schedules·Webhooks·100s of pre-built events·Custom integration cron triggers·Bulk runs across datasets·Call via API

Channels & sharing

Voice agents·Embeddable widgets & chat bubble·Public & marketplace listing·Shareable links·Welcome & starter messages·Suggested replies

Oversight & quality

Email & Slack escalations·Condition to action alerts·External message relay·PII data masking·Pre-publish eval sets & thresholds·Draft, active & pinned versions·Production sampling & alerts

Multi-agent orchestration

Node-and-edge graphs·Agent, tool & condition nodes·Handoffs & sub-agent calls·Parallel & branching runs·Visual drag & drop builder

Experts and engineers build together

Three ways to build. Every path ships to the same runtime, gated by the same evals.

Drag and drop

Compose agents and workforces on a visual canvas. Drag in agents, tools, and approvals in a flow anyone on the team can read.

Drag and drop

Compose agents and workforces on a visual canvas. Drag in agents, tools, and approvals in a flow anyone on the team can read.

Build with AI

Describe the agent in plain language and Invent builds it: the prompt, the tools, and the evals to prove it works.

Build with AI

Describe the agent in plain language and Invent builds it: the prompt, the tools, and the evals to prove it works.

Build with MCP

Engineers drive the same platform from Claude Code, Codex, or Cursor. Create agents, link knowledge, and run evals over MCP.

Build with MCP

Engineers drive the same platform from Claude Code, Codex, or Cursor. Create agents, link knowledge, and run evals over MCP.

Evals, so you know your agents work

Every agent ships with a test suite built from your real cases, run on every change and sampled in production.

Define what good looks like

Invent drafts the suite from past tickets, calls, and records. Add your own cases in plain language.

Define what good looks like

Invent drafts the suite from past tickets, calls, and records. Add your own cases in plain language.

A failing version never ships

Every change re-runs the suite before it publishes. Fail an eval and the version is blocked.

A failing version never ships

Every change re-runs the suite before it publishes. Fail an eval and the version is blocked.

Catch drift before customers do

Relevance keeps scoring a sample of production runs and flags dips the moment they start.

Catch drift before customers do

Relevance keeps scoring a sample of production runs and flags dips the moment they start.

An n8n-style workflow builder, built in

When a task always runs the same way, you shouldn't pay a model to re-figure it out each run. Chain vendor actions, branches and loops, and call a model only for the step that needs it.

Deal intakeSaved

Add node

Connect to 1,000+ apps

Browse all integrations

Triggers that can run your business

Start an agent from a schedule, an event in your apps, a webhook, or a custom signal. Then, when a thousand tasks land at once, a managed queue paces the work so nothing downstream falls over.

Schedule

Every week, day or hour

Your apps

Events from connected apps

Webhook or API

Trigger agents from anywhere

Custom signals

Build custom data signals

Every weekday, 9:00amSchedule

New Salesforce leadYour apps

order.createdWebhook

Enrichment completeSignal

HubSpot deal updatedYour apps

payment.failedWebhook

Inbound support emailYour apps

CSV row importedSignal

Every weekday, 9:00amSchedule

New Salesforce leadYour apps

order.createdWebhook

Enrichment completeSignal

HubSpot deal updatedYour apps

payment.failedWebhook

Inbound support emailYour apps

CSV row importedSignal

Without Relevance

Rate limit

Calls get rejected and runs fail silently. You build and run the queue yourself.

Relevance

Managed queue

Queue / brokerWorkers + autoscaleConcurrency capsRetries + backoffDead-letter queueWork-hours pacing

Built for enterprise teams

Run agents on your real data with the access controls, audit trails, and residency guarantees enterprises require all built in.

SOC 2

GDPR

Security & data privacy

Data residency
PII masking
Audit logs
No training on your data

Access & controls

Role-based access control
SSO / SAML
Human-in-the-loop approvals
Version control

Monitoring & oversight

Real-time monitoring
Full agent tracing
Cost visibility
OTEL & Delta Share export

For teams who own the number

Building an agent is easy. Relevance is built for what comes next: your experts owning quality, operations, and outcomes at scale.

Who manages agents day-to-day?

Your ops and revenue teams — the people closest to the work. We have dashboards and tooling designed specifically for agent management.

Your engineers — agents are configured via API, SDK, or CLI. The Claude Console adds GUI tracing and admin controls.

Who manages agents day-to-day?

Your ops and revenue teams — the people closest to the work. We have dashboards and tooling designed specifically for agent management.

Your engineers — agents are configured via API, SDK, or CLI. The Claude Console adds GUI tracing and admin controls.

Who owns agent quality?

Domain experts, with pre-deployment scenarios, production checks, and monitoring they can manage themselves.

Your engineers, via runtime self-healing (Outcomes) plus a Console eval tool for pre-deployment prompt testing.

Who owns agent quality?

Domain experts, with pre-deployment scenarios, production checks, and monitoring they can manage themselves.

Your engineers, via runtime self-healing (Outcomes) plus a Console eval tool for pre-deployment prompt testing.

How do agents receive tasks?

Native triggers for CRMs, email, calendar, and more. Plus webhooks, cron, and the ability to build custom triggers in platform.

Native scheduled deployments (cron) and webhooks, plus on-demand runs — all in beta.

How do agents receive tasks?

Native triggers for CRMs, email, calendar, and more. Plus webhooks, cron, and the ability to build custom triggers in platform.

Native scheduled deployments (cron) and webhooks, plus on-demand runs — all in beta.

What can my agents access?

1,000+ native connectors plus MCP support and the ability to build custom connectors.

Built-in hosted tools (web, code, files) plus MCP and Anthropic’s Connectors Directory — MCP-based, not native connectors.

What can my agents access?

1,000+ native connectors plus MCP support and the ability to build custom connectors.

Built-in hosted tools (web, code, files) plus MCP and Anthropic’s Connectors Directory — MCP-based, not native connectors.

How much can I tune performance?

Full orchestration with parallel streams, deep nesting, evals, and production monitoring — all GA.

Strong model quality, but single-level orchestration: a coordinator delegating to up to 20 agents, no nested sub-delegation.

How much can I tune performance?

Full orchestration with parallel streams, deep nesting, evals, and production monitoring — all GA.

Strong model quality, but single-level orchestration: a coordinator delegating to up to 20 agents, no nested sub-delegation.

How much can I control cost?

Full. Route to any model provider and optimize cost per agent, per task.

Claude models only — no routing to third-party providers. You can down-tier within the Claude family (Haiku vs Opus).

How much can I control cost?

Full. Route to any model provider and optimize cost per agent, per task.

Claude models only — no routing to third-party providers. You can down-tier within the Claude family (Haiku vs Opus).

How do I govern this?

Single pane for access control, permissions, and data security. Lots of control over approvals & escalations.

Tool-level permissions plus Console RBAC, SCIM, per-user spend limits, and an Admin API.

How do I govern this?

Single pane for access control, permissions, and data security. Lots of control over approvals & escalations.

Tool-level permissions plus Console RBAC, SCIM, per-user spend limits, and an Admin API.

"The ability to be vendor agnostic and the ability to scale across a breadth of functions is a really key feature."

Levi Watters

Partner, KPMG Australia

"The key for us was how we can modularize industry knowledge and the best playbooks, and apply it."

Allen Roh

Senior Marketing Manager, Autodesk

"We're looking for every place where AI can allow sellers and customer success reps to be more engaged with customers."

Rob Giglio

Chief Customer Officer, Canva

"The ability to be vendor agnostic and the ability to scale across a breadth of functions is a really key feature."

Levi Watters

Partner, KPMG Australia

Agents@Work

Case study

Agents@Work

Case study

Case study