Product & User Interface
Where people and systems meet the harness: chat interfaces, editor plugins, APIs and reviewer dashboards for enterprise AI applications. Every surface runs on the same controls underneath.
What is this layer about
A harness is only useful if people can reach it. The product and interface layer is every surface through which employees, developers, reviewers and other software talk to AI: chat interfaces, IDE integrations, public and internal APIs, SDKs, and the dashboards that administrators use to run it all.
The key idea is one set of controls behind many front doors. Whether someone asks a question in a chat window, a developer calls an endpoint from a service, or an agent is embedded inside a business application, the same identity, policy, routing and audit apply. The interface changes; the control does not.
This layer also carries the human side of oversight. Approval inboxes, review queues and agent dashboards put a person in the loop where a decision needs one, with the context to decide quickly and a clear record of what they decided.
What do we deliver
Chat and assistant interfaces
Governed chat experiences for employees, with streamed responses, conversation history, file attachments and role-aware access to the models, tools and data each person is allowed to use.
IDE and developer integrations
Assistants and agents inside the editors developers already use, connected to approved models and tools, so coding help follows the same policies as every other AI request.
APIs and SDKs
Documented public and internal APIs with client libraries, so product teams can add AI to their own applications through one governed entry point instead of calling providers directly.
Agent dashboards and approval inboxes
Views for administrators and reviewers to monitor running agents, inspect activity, pause or stop work, and approve or reject sensitive actions with the full context in front of them.
SSO and role-based experiences
Sign-in through the enterprise identity provider, with what each person sees and can do shaped by their role, team and the sensitivity of the data involved.
Embedding AI in business apps
Patterns for placing AI inside the tools people already work in, such as collaboration apps, internal portals and line-of-business systems, rather than asking them to adopt a new destination.
Technology stack
- Chat and web front ends
- React
- Next.js
- Vercel AI SDK
- assistant-ui
- Open WebUI
- IDE and agent integration
- VS Code extension API
- JetBrains Platform
- Language Server Protocol
- Model Context Protocol
- Agent Client Protocol
- APIs, SDKs and streaming
- OpenAPI
- REST
- GraphQL
- gRPC
- Server-Sent Events
- WebSockets
- Identity and access
- OpenID Connect
- SAML
- OAuth 2.0
- Microsoft Entra ID
- Okta
- Keycloak
- Dashboards and embedded surfaces
- Grafana
- Streamlit
- Retool
- Slack
- Microsoft Teams
These are the kinds of technologies this layer is typically built from; the choice of tools stays with the customer.
Business outcome
AI where employees already work. People use governed AI inside the chat tools, editors and business applications they already know, which makes adoption easier and keeps usage within approved channels.
Consistent control on every surface. Because each interface sits on the same harness, identity, policy and audit apply the same way whether the request comes from a chat, an IDE or an API.
Faster product delivery for builders. Product teams add AI features through one documented, governed API instead of building provider integrations and safeguards separately for every application.
Humans in the loop, with context. Reviewers and administrators can see what agents are doing and decide on sensitive actions quickly, with a clear record of who approved what.
Frequently asked questions
What does the product and user interface layer cover?
It covers the surfaces people and applications use to reach AI: chat, IDE plugins, business applications, APIs, admin consoles and review screens. Each one connects to the same harness underneath, so the controls stay consistent.
Can different roles see and do different things?
Yes. The layer relies on your identity provider, so access can follow role. A reviewer, an administrator and an end user can each be given only the screens, actions and data their role allows.
How do human review screens fit in?
When an agent reaches a sensitive step, it can pause and show a reviewer the context. The reviewer approves, edits or rejects, and the decision is recorded, so people stay in control of outcomes that matter.
How does this layer relate to the layers below it?
It is the visible top of the stack. Requests pass down through safety and policy controls, runtime, context and model layers, so an interface never reaches a model directly, and every surface is governed the same way.
Questions about your AI rollout?
Talk to us about how the seven layers fit your models and data.