Roots AI, explained
Roots AI is now Bevaya. Here is what changed, and where the API model stops.
Direct answer · verified 2026-06-16
Roots AI is the insurance AI company Roots Automation, Inc. (the domain is roots.ai). On May 28, 2026 it launched a new flagship platform, Bevaya, and rebranded the product to that name. The legal entity is still Roots Automation. Visit roots.ai/platform today and you get a 301 redirect to bevaya.ai.
First, which "Roots AI" do you mean?
The name maps to three unrelated things. Most people searching it want the first one, but it is worth being sure before you read a comparison or send a vendor email to the wrong company.
Roots Automation
Insurance AI at roots.ai, now branded Bevaya. Founded 2018 by former AIG executives. The one almost everyone means.
Root AI (robotics)
A greenhouse-harvesting robotics startup. Its Virgo robot was acquired by AppHarvest in April 2021 for about $60M. Unrelated to insurance.
RootsAI Foundation
A nonprofit working on AI bias and endangered-language preservation. A community effort, not a vendor.
The rest of this page is about the first one: Roots Automation and its Bevaya platform, because that is the one people compare against desktop automation.
What Roots Automation / Bevaya actually does
Bevaya is a cloud platform for carriers, brokers, and TPAs to design, deploy, and govern AI agents across three areas of insurance work:
- Underwriting: submission intake, ACORD form extraction, loss run processing, renewal handling.
- Claims: FNOL/FROI setup, claims indexing, legal demand extraction, medical bill processing.
- Policy servicing: endorsement processing, certificate of insurance creation, premium audit.
It runs on InsurGPT, a model the company says is trained on over 300 million proprietary insurance documents, and it reaches your systems of record through APIs, SFTP, and direct connections to platforms like Guidewire and Salesforce. By its own account Roots Automation has put 115+ workflows into production, including at three of the top five P&C carriers, and it raised a $22.2M round before the rebrand. That is a real, credible insurance automation company. None of what follows is a knock on it.
The anchor fact: the redirect, and what it implies
You can confirm the rebrand yourself in two seconds. The product path on the old domain now answers with a permanent redirect to the new one:
The interesting word on the Bevaya platform page is not in the marketing copy, it is in the integration line: APIs, SFTP, and direct connections. That is the right architecture when your systems of record can be connected to. It is also the exact place the model runs out of road. A cloud platform needs something on the other end to call. If your claims live on a mainframe green screen or an old policy admin desktop with no API and no clean feed, there is nothing to connect to.
Where Mediar fits: the desktop with no API
Mediar is not an insurance platform and not a document model. It is a Windows desktop agent that watches a workflow once, then executes it through the operating system's accessibility tree, the same interface a screen reader uses. That means it can drive an application that exposes no API at all: it reads the fields and buttons the app already publishes to assistive technology, and types into them the way a person would.
Two ways to reach a legacy claims system
Because there are no pixel matchers and no recorded selectors, the agent self-heals when a field is relabeled or a button moves in a quarterly UI refresh. And because it is open at the edges, teams can extend it with the Terminator SDK at github.com/mediar-ai/terminator.
Roots AI / Bevaya vs Mediar, side by side
These are not the same tool, and the honest framing is which layer your problem lives in, not which is "better." If your data already moves through modern APIs, a cloud insurance platform is the cleaner fit. If the last mile is a person retyping into a desktop nothing can connect to, that is the Mediar layer.
| Feature | Roots AI / Bevaya | Mediar |
|---|---|---|
| What it is built for | Insurance only: underwriting, claims, policy servicing for carriers, brokers, TPAs | Any repetitive workflow on a legacy Windows desktop, across insurance, banking, healthcare, F&B, manufacturing |
| How it reaches your data | Cloud SaaS connecting via APIs, SFTP, and direct connections to systems like Guidewire and Salesforce | Windows accessibility APIs, the same tree a screen reader reads, so it works on apps that expose no API at all |
| Core model | InsurGPT, trained on 300M+ proprietary insurance documents, for reading and extracting documents | An agent that watches a workflow once, then executes the clicks and keystrokes a person would |
| Green-screen and mainframe claims systems | Needs an API, SFTP, or direct connection to ingest and write back | Drives the screen directly when there is nothing to connect to |
| When the UI changes | Document and API mappings re-validated by the vendor | Self-heals on relabeled fields and moved buttons because there are no pixel matchers or selectors |
| Extensibility | Workflow Canvas and pre-built agents inside the platform | Open-source Terminator SDK at github.com/mediar-ai/terminator |
| Pricing shape | Enterprise SaaS contract (not published) | $0.75 per minute of runtime, $10,000 turn-key program fee that converts to credits, no per-seat licensing |
Roots Automation / Bevaya is an insurance-specific cloud platform; Mediar is a general desktop-execution layer. Choose by where your bottleneck actually sits.
“Claims intake at one mid-market carrier dropped from 30 minutes per claim to 2 minutes once the agent drove the claims desktop directly. That is roughly $750K a year, and that math is the AP team's headcount, not a press-release number.”
Mediar insurance claims deployment
How to decide in one question
Ask where the work actually stalls. If the answer is "we cannot read the documents fast enough," an insurance document platform like Bevaya is built for that. If the answer is "the documents are fine, but someone still spends half their day retyping into a system we cannot get an API for," that is the desktop layer, and that is what Mediar runs. The two are not mutually exclusive; plenty of teams keep their document AI and bolt desktop execution onto the no-API systems behind it.
Bring the workflow Roots AI cannot reach
Book a call and we will run an agent against the legacy desktop system that has no API, and show you the before and after.
Frequently asked questions
Frequently asked questions
What is Roots AI?
Roots AI is the AI insurance company Roots Automation, Inc. (the domain is roots.ai). Founded in 2018 by former AIG executives, it builds agentic AI for carriers, brokers, and TPAs. On May 28, 2026 it launched a new flagship platform called Bevaya, which is now the public face of the product; roots.ai/platform redirects to bevaya.ai.
Is Roots AI the same as Bevaya?
Yes. Bevaya is the new platform name from Roots Automation, Inc. The legal entity stays Roots Automation, but the product rebranded to Bevaya at launch on May 28, 2026. If you visit roots.ai/platform you get a 301 redirect to bevaya.ai/platform.
Is Roots AI the harvesting robot company?
No, that is a different company. Root AI was an agricultural robotics startup whose Virgo harvesting robot was acquired by indoor farming company AppHarvest in April 2021 for roughly $60 million. It is unrelated to Roots Automation's insurance platform.
What does Roots Automation's platform actually do?
It automates insurance document and decision work: submission intake and ACORD extraction in underwriting, FNOL/FROI setup and claims indexing in claims, and endorsement and COI processing in policy servicing. It runs as cloud SaaS powered by InsurGPT and connects to existing systems through APIs, SFTP, and direct connections.
How is Mediar different from Roots AI / Bevaya?
Roots/Bevaya is a cloud platform that reads documents and connects to insurance systems through APIs. Mediar runs on the Windows desktop and drives applications through accessibility APIs, so it works on the legacy claims and policy systems that have no API for a cloud tool to call, including mainframe green-screens. Mediar is also not insurance-specific.
Where does the API-based approach hit a wall?
When the system of record has no API and no clean SFTP feed. Plenty of regional carriers run claims on a mainframe terminal or an old policy admin desktop. A cloud platform needs an integration point that does not exist there. That gap is the reason the accessibility-API approach exists.
Can the two be used together?
They solve different layers. If your data already flows through modern APIs, a cloud platform is a clean fit. If the last mile is a person retyping into a desktop app that nothing can connect to, that is where Mediar runs. Some teams keep their document AI and add desktop execution for the no-API systems.
Keep reading
Insurance claims intake automation
Driving Guidewire ClaimCenter and Duck Creek Claims directly, not just parsing ACORDs.
Insurance automation, Mediar solution
From submission to adjuster assignment in under 2 minutes on legacy claims systems.
Enterprise process automation
Where RPA stalls on legacy desktop apps and what the accessibility-API approach changes.
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.