Arc · AI Resource & Control

The control plane for every AI agent you run.

Arc finds the agents running across your clouds, puts each one through evaluation before it goes live, decides which tools it may call, and can stop any of them in seconds.

Support agent Sales agent Ops agentdrift Finance agent HR agent
  • Active
  • Discovered
  • Drifting
  • Paused
Amazon Bedrock AgentCoreAzure AI FoundryGoogle Vertex AIModel Context ProtocolAgent2Agent cardsOpenTelemetry Amazon Bedrock AgentCoreAzure AI FoundryGoogle Vertex AIModel Context ProtocolAgent2Agent cardsOpenTelemetry
0clouds discovered, one registry
0lifecycle stages, one vocabulary
0checks before any agent goes live
0to pause an agent and stop its billing

See it work

Six things Arc does, in motion

01

It finds every agent

Connect a cloud account with one scoped role. Arc discovers the agents running there, across AWS, Azure and Google Cloud, and keeps one registry in step with all three.

  • Nothing it finds is presented as in service
  • Each agent gets an owner before it gets a status
AWS Azure Google Cloud registry
02

Going live is a gate, not a default

An agent becomes active only when its record is complete, its guide is written, the safety suite passes against the live agent, its own quality bar is met and no policy blocks it.

  • The checks run again every time someone presses Activate
  • Each failure says exactly what to fix
Agentregistered Guide Safety Quality Policy Active registered → active
03

Arc decides. The gateway enforces.

Agents reach business systems through one door. An agent proves who it is with its own cloud role, and can call only the tools it was granted.

  • No credentials inside agents
  • Every refusal lands in Arc
  • If Arc is down, the gateway keeps enforcing
Agentown role Gatechecks Arc Salesforceread ServiceNowread refused: not granted
04

A kill switch that means it

Pause an agent and Arc saves its runtime, removes it in the cloud, ends its sessions, cuts its access to tools and stops its compute bill. Resume brings it back.

Try it: the picture is live.

Callersusers, flows Agentrunning Model Tools compute billing paused · billing stopped
05

It catches what nobody signed off on

What an agent declares it uses, against what its traces show it calling. A tool wired in after go-live shows up as drift, with the call count and the day it was last seen.

  • Compared continuously, not once at launch
  • Counts gateway calls as well as traces
declared in its guide search_incidentsget_runbookget_graph Agent search_incidents get_runbook delete_record ! drift: called, never declared
06

Every dollar has an agent behind it

Usage, tokens and cost come from each agent's own traces, priced and rolled up by Arc: by agent, by model, by tool and by the workflow that spent it.

  • Full traces for active agents, counts for the rest
  • Value models turn usage into return, with the confidence stated
Support Sales Ops Model Areasoning Model Bfast cost / 30d$4.2k sample data

The rest of the platform

From first sighting to retirement

Evaluations

Safety and quality suites run against the live agent, with checks, scores, grades and a review queue.

A

Governance

Policies as data, checked across every asset. Findings say what is wrong, and with what.

Connections map

Dependencies, blast radius and orphaned assets, as a graph of the estate.

A2A agent cards

How each agent describes itself: skills, examples and what it changes.

Value & ROI

Value models and a rate card turn what agents do into return over a measured scope.

Solutions

Business-facing entry points that run registered agents in order, with people approving where it matters.

How it works

Connect, govern, enforce

  1. 01

    Connect your clouds

    An onboarding script creates one scoped, read-mostly role per account. Arc discovers what runs there and keeps it in sync.

  2. 02

    Govern every agent

    Owners register their agents, upload a guide and run the evaluations. Going live is a decision someone makes.

  3. 03

    Enforce at runtime

    The gateway allows only granted tools, drift flags the unexpected, and the kill switch stops any agent in seconds.

Also in Arc 360

The AI Command Center

Adoption of AI coding assistants, enablement and certifications, efficiency uplift by team and project, and licence spend you can reclaim.

Usage metricsEnablementEfficiencySteering governanceLicence FinOps

Built for security review

Controls an auditor can read

Least privilege by default

One scoped role per cloud account. Each agent runs as its own role and holds no credentials for the systems it uses.

Every tool call checked

Agents reach business systems through one gateway. Only the tools an agent was granted get through; every refusal is recorded.

An audit trail for every decision

Who paused an agent and why, who granted a tool and when, every publish to the gateway and every refused call.

Role-based access

Viewers read, operators pause, contributors register, administrators govern. Checked on the server on every request.

Take control of your agent estate.

Open the Arc Control Plane and see every agent you run, and what each one is allowed to do.

Open the Control Plane