Port of Context for Insurers
Agents need your firm’s risk context before they can act on it.
Port of Context builds agent execution infrastructure for insurers. Relationships, policy systems, and decades of underwriting judgment, made legible to AI. That takes engineering, not vibe-coding.
Your ontology[ meaning ]
How your book’s data interconnects, so that agents read risk the way your underwriters do.
Your taxonomy[ structure ]
How your firm classifies its book: lines, classes, treaties, in its own terms.
Your deep data[ source ]
Decades of your firm’s most sensitive records, down to every submission, loss run, and claim file.
Live at Prudential Financial and Block.
A $7B investment firm runs 50+ custom agents on Port of Context.
Underwriting and claims
Triage every submission and claim without drowning in attachments.
Every submission and claim arrives as a stack of emails, PDFs, and loss runs. Port of Context engineers agents that read all of it before your team opens a file.
Policy and treaty archive
Every policy and treaty you’ve ever written—cleaned up, sorted, and queryable.
Policy wordings, endorsements, treaties, and claims files, scattered across decades of underwriting. Port of Context reads every file and sorts it into one structured index your team can query.
Put stress-tested agents on top of the tools you already use.
Don’t force new workflows and UI onto your team. Run agents on the tools your team is already used to.
How it works
Engineers build the foundation. Your team builds on top.
Resilient agents come from disciplined foundations.
- 1
Clean the data at its source
We index all of your deep data so that it can be accessed by agents with proper controls and tooling.
- 2
Build a custom MCP server over it
We tune the server the exact way agents need to read and query it.
- 3
Eval rigorously
We build evaluation cases on your actual workflows and keep refining them, so each agent proves it executes properly before it ever reaches production.
- 4
Your firm builds agents in one prompt
With the MCP server in place, your own team can now reliably build production-grade agents on top.
Map Port of Context to your environment.
We'll look at your infrastructure, models, and compliance constraints, then tell you honestly whether Port of Context fits. A real engineer follows up fast.
FAQs
- What is the difference between an ontology and a taxonomy?
- A taxonomy is how a firm categorizes its data in its own terms. It tells an agent where things belong. An ontology defines how those categories relate to each other and encodes the specific logic the firm uses to put its data to work. An agent needs both. The taxonomy tells it where to look, and the ontology tells it what the data means once it gets there.
- What is agent execution infrastructure?
- Agent execution infrastructure is the layer that lets AI agents run on a firm’s systems and data in production. It covers how agents read and query the data, which tools they can call, how their access is controlled, and how every run is traced back to its sources. Port of Context builds this layer for regulated firms on top of the tools they already use.
- How do AI agents handle claims files and policy documents?
- Badly, unless the data is prepared first. A claims file arrives as adjuster notes, medical records, correspondence, and policy wordings with no shared structure, and an agent pointed at that raw pile will miss things. The reliable approach is to index the material at its source, build a custom MCP server over it, and give agents controlled, queryable access. Every answer then traces back to the underlying document, so your team can verify what the agent found before it reaches an underwriting or claims decision.
- Are we locked into one AI model or vendor?
- No. The infrastructure is model-agnostic. Agents run on Claude, GPT, Gemini, or open-weight models, and the firm can switch when a provider changes pricing, degrades, or gets ruled out by policy. The data layer, the MCP server, and the ontology work stay yours, so a model change never means rebuilding the foundation.
- Do we need to replace the tools our team already uses?
- No. We build on top of the tools your firm already runs, including Excel, Outlook, SharePoint, Salesforce, and core systems such as Guidewire and Duck Creek. Agents reach those systems through the MCP server, with access controls in place. Your team keeps working where it already works, and the agents operate on the same data.
- What does it take for our own team to build an agent?
- A prompt, once the foundation is in place. The engineering happens up front, in cleaning the data at its source, building the MCP server, and writing evaluation cases against your actual workflows. After that, someone on your team describes the agent they need and it runs on infrastructure that is already controlled and traceable.