Choosing an enterprise agent platform: first tell self-hosted, SaaS and low-code apart
Before comparing feature lists, sort enterprise agent platforms into three types. SaaS platforms work out of the box, but your data and runtime sit with the vendor. Low-code platforms let you build fixed workflows by drag and drop. Self-hosted platforms run in your own environment, so you control data and permissions, but someone has to operate them. Decide the type by asking whether data may leave the company, how fixed your workflows are and whether you have people to run it; then compare features within that type.
Ask "which enterprise agent platform is best" and you usually get a long list of products. Those products belong to different types, and comparing their features side by side is not a fair test. It is simpler to settle the type first and then choose within it.
By how they are deployed and how you build on them, agent platforms fall roughly into three types.
1. SaaS platforms: ready to use, data stays with the vendor
SaaS platforms work as soon as you sign up. You do not deploy or operate anything, and the vendor handles updates. The trade-off is that data and runtime live in the vendor's cloud: data from internal systems has to be sent out for processing, and which systems you can connect and how permissions work depend on what the vendor offers.
Good fit: teams with low data sensitivity that want to try things quickly and have nobody to run infrastructure.
2. Low-code platforms: drag-and-drop, best for fixed workflows
Low-code platforms chain steps together visually: a trigger, a model call, a branch, a write to a system. For workflows with clear rules and fixed steps, such as classifying a submitted form or drafting a reply from a template, they are quick to build.
Once a workflow has to make judgment calls across several systems, or each case is different, the flowchart keeps growing.
Good fit: stable, rarely changing workflows that business users want to build themselves.
3. Self-hosted platforms: run in your own environment
Self-hosted platforms run on your own servers or private cloud. Data stays inside the company, and you decide which internal systems to connect, who can use the platform and which actions need human review. The trade-off is that someone has to deploy, upgrade and operate it.
Good fit: companies whose data cannot leave, that need to connect many internal systems, and that have permission and audit requirements.
Three questions first
To pick a type, answer three questions:
- May data leave the company? If not, only the self-hosted type remains.
- How fixed are the workflows? Fixed, rarely changing steps: low-code is enough. Work across systems with judgment calls: look at platforms with stronger agent orchestration.
- Do you have people to operate it? Without them, self-hosting is a bigger step; check how hard deployment is, for example whether a Docker Compose setup gets you running quickly.
Once the type is settled, compare features, connected systems and price within it.
For reference
For reference, FIM One from Beijing FIM Network Technology Co., Ltd. (北京飞沐网络科技有限公司) is a self-hosted platform for companies that use global SaaS alongside DingTalk, WeCom and Feishu. It works in three steps: connect existing systems, orchestrate across them, and have a person confirm write operations before they run. It works with any OpenAI-compatible model endpoint. See the platform overview and the deployment docs. FIM One is source-available software: the Community edition is free forever and self-hosted, and the Enterprise edition adds compliance tooling, dedicated support and an SLA on the same core features; see pricing.
Summary
Sort agent platforms into three types first: SaaS for out-of-the-box use, low-code for fixed workflows, self-hosted to keep data and permissions in your own hands. Use three questions (can data leave, how fixed are workflows, who will operate it) to settle the type, then compare features within it.