Agent-based fits when
The data sits on laptops or servers that are often off the network, or on systems with no interface you can read remotely.
Category comparison. Data discovery.
What each approach asks of your hosts, your network and your team, shown on one database discovered both ways.
By Dipal Panchal, co-founder and CEO · Updated November 8, 2026
Agent-based discovery installs a program, the agent, on each server, endpoint or virtual machine. The agent reads data where it sits and reports to a central service. Agentless discovery installs nothing on the source: a scanner elsewhere connects through the source’s own interfaces, using an account you create for it.
Choose per source. Use agents where a machine leaves the network or offers no remote interface, and a connection where the source offers one. NIST described the same trade-off for patch management tools in SP 800-40 Revision 3, a guide since withdrawn.
Cells describe the typical tool in each approach. Check your own tool against them.
| Aspect | Agent-based | Agentless |
|---|---|---|
| Acts on | Servers, endpoints and virtual machines that run the agent | Databases, file shares, cloud storage and SaaS apps that offer an interface |
| Main strength | Reaches local disks and machines that are rarely on the network | Nothing to install or patch on each source |
| What it cannot do | Cover appliances and platforms the agent doesn’t support, or hosts where you can’t install software | Reach a source behind a firewall or address translation, or a machine that is off the network |
| Best fit | Laptops, and servers with no remote interface | Estates made mostly of databases, file shares and cloud services |
| A typical scenario | Find files on a field laptop that rarely connects | Find card numbers in a production database through a read-only account |
| Where it runs | On every host in scope | On a scanner host that connects to each source |
| What you grant | Rights to install software, often with administrator privileges on each host | An account on each source with the read rights the scan needs, and a network path to it |
| Load on the source | Processor and memory on each host while the agent works | Queries and reads against the source during a scan, plus network traffic |
| Upkeep | Agent updates, compatibility checks and resources on every machine | Accounts, credentials and network rules for the scanner |
| What a security review asks | What the agent can do on each host, and how it is updated | Where the scanner runs, what each account can read and how credentials are stored |
Suppose a PostgreSQL server holds a billing database in a locked-down network segment. You need to know which tables hold card numbers.
| Step | Agent-based | Agentless |
|---|---|---|
| What is installed | An agent on the database host, approved and installed by a server administrator | Nothing on the host |
| What is granted | Install rights on the host, then the agent’s access to local files and the database | A read-only database account created for the scanner |
| Network path | The agent calls out to a central service, so the segment needs an outbound rule | The scanner reaches the database port, so the segment needs an inbound rule from one host |
| Load while it runs | Processor and memory on the database host, shared with the database | Queries against the database. IRIS is designed to limit load on production sources. |
| Upkeep next quarter | Agent update, tested against the host’s patch cycle | Credential rotation and the firewall rule |
| When it breaks | A host patch stops the agent; its data goes quiet until someone notices | The account expires; the connection reports failed and the source shows as unread |
The data sits on laptops or servers that are often off the network, or on systems with no interface you can read remotely.
The data sits in databases, file shares, cloud storage and SaaS apps that already offer a way to read them with an account.
Your estate mixes endpoints with databases and cloud services. Choose per source, and record which approach covers which data.
Pick one production database and one file share, and put the same questions to each approach.
Read-only access, deployable in your environment, agentless where the source supports it.
IRIS uses read-only access to your sources, agentless where the source supports it, with accounts you create.
Each source connection is tested and reports connected, partial or failed.
IRIS is designed to limit load on production sources.
IRIS can be deployed in your own infrastructure, on-premises or in any cloud, so your data stays within your environment.
Capabilities vary by edition.
What IRIS reads, what it keeps and how each scan runs, from connection to evidence.
Each component, where it runs, what it reads and what it stores.
The sources IRIS connects to, and what it reads from each one.
No. Where the source supports it, IRIS connects without an agent, read-only, using an account you create for it.
No, not by default. It moves the risk from software on every host to one scanner and its accounts. Review where the scanner runs, what each account can read and how credentials are stored.
Yes, some do. Ask which approach each of your sources would use, and what the tool can read through it.
A host in your cloud account or data center, a read-only account for each source, and network access from IRIS to those sources. The host needs a time source, which can be your own, and a connection to the license service once, at activation.
Source credentials are kept in the IRIS database inside your deployment, encrypted at rest. Give IRIS a dedicated read-only account for each source.
Run the access test on one production database this week, then book a demo or take the tour.