Your AI agents are already touching production. Are they governed?
Exemplar is the policy layer that governs what agents can do — token budgets, approval gates, and a full audit trail across every model, tool, and framework.
token_spend.week
$0
calls.blocked.today
0
model_routing.savings
0%
audit_trail.completeness
100%
How it works
Every agent call passes through
the same policy surface.
The same visuals that power Exemplar Console — Relay's decision wire, the living MCP gateway, and guardrails — live here so you can see the control plane before you sign in.
Marshal
Every tool. One endpoint.
Marshal serves your connectors, memory, prompts, and skills over MCP — any agent or IDE can use all of them the moment it connects.
Marshal offerings
View all →
The platform
Marshal, Relay, and the rest of the control plane.
Exemplar is a suite — not a single tool. Govern agents in the IDE, in production workflows, and across your Dev connectors from one policy surface.
Marshal
Agent harness with the primitive tools to run agents in production
Tools, prompts, skills, gateway, evals, guardrails, and HITL — the primitives production agents need.
Relay
Hooks that govern Cursor and Claude Code
Control plane for Cursor and Claude Code hooks — allow, ask, or deny before risky actions land.
DevX Assist
Governed MCP for Dev connectors & integrations
Connect Dev integrations and expose them as governed MCP — same policy and audit as the console.
The problem
Your agents are running in production.
Nobody knows what they're spending.
No budget limits. No access controls. No audit trail. Most teams discover the problem after it becomes an incident.
3.2×
average token overspend when agents run without model routing or budget enforcement
74%
of agent failures traced to unrestricted tool access — agents calling APIs they should never reach
0
audit records kept by most teams — no log of what ran, what it touched, or who approved it
Before vs. after
The same agent. Two very different outcomes.
One runs ungoverned — expensive, unrestricted, unaccountable. The other runs through Exemplar.
Without Exemplar
With Exemplar
The numbers
What governed agents actually look like.
Data from engineering teams running AI agents through Exemplar — before and after governance was enabled.
Weekly token spend
before vs. after Exemplar
Model routing distribution
tasks auto-matched to right model
Blocked calls by category
last 30 days across all agents
Agent run outcomes — daily
passed / approved / blocked
Guardrails
Define exactly what your agents
can and cannot touch.
Set access policies once. Every agent — on every framework — operates within them.
Tool access control
agent-code-review · last run 4m ago
Cost governance
budget tracking · all agents · this week
The autonomous agent era is here.
Run it on Exemplar.
Marshal for production agents. Relay for Cursor and Claude Code. Studio and DevX Assist for the rest of the stack — one policy surface.
Blog
From the team
AI & platform
What Is Loop Engineering? The Complete Guide
Loop engineering is the discipline of designing the plan-act-observe cycle that lets an AI agent complete multi-step work: termination conditions, state, retries, and cost bounds. What it is, how it differs from harness engineering, and how to build one.
ReadAI & platform
Loop Engineering vs Harness Engineering: What's the Difference?
Loop engineering designs the plan-act-observe cycle an agent runs. Harness engineering governs what that loop is allowed to do. Clear definitions, a side-by-side comparison, and which to build first.
ReadAI & platform
Loop Engineering: 25 Questions Answered
Every question engineering teams ask about loop engineering — answered directly. What it is, how to design termination and retries, how it differs from harness engineering, cost control, and safety.
ReadAI & platform
The Loop Engineering Checklist: 12 Things Before You Ship a Standing Agent Loop
A practical checklist for teams shipping standing AI agent loops: termination conditions, retry design, cost bounds, safety gates, and monitoring — the 12 things to put in place before a loop runs unattended.
ReadAI & platform
Best AI Agent Loop Orchestration & Control Tools in 2026
The best tools for building and running AI agent loops in production — orchestration frameworks, durable execution engines, and the governance layer that keeps standing loops safe and bounded. Compared and ranked.
ReadAI & platform
Best AI Agent Governance Platforms in 2026
The best AI agent governance platforms for controlling what AI agents can do in production — policy gates, approval workflows, token budgets, and audit trails. Compared and ranked for engineering teams.
Read