Which AI models do you use?

Whichever fits the job. We work model-agnostic through the Vercel AI Gateway, so we can use Claude, GPT, or Gemini and switch between them without rewriting your app.

Can you add AI to our existing product?

Yes. Most of our AI work is features wired into an app that already exists, behind clean interfaces so it is easy to change later.

How do you keep it from making things up?

Retrieval over your own data, structured output, guardrails, and human review where errors are costly. We also measure quality with evaluations rather than guessing.

What does it cost to run?

It depends on usage and the model. We build with cost controls and right-sized models, and we can pass token costs through transparently or bundle them, whichever you prefer.

Is this the same as your AI Automation service?

Related. This page is about building AI into products. AI Automation is about removing repetitive work in your operations. They use the same tools and often run together.

Do we own what you build?

Completely. The code, the prompts, and the configuration are yours.

Platform capability

We build with large language models.

Claude, GPT, and Gemini, reached through a single gateway, so your product uses the best model for each job and can switch when the market moves. Chat, copilots, extraction, classification, and agents that actually do work.

In production

Talk about an AI build

Can Foundri build AI and LLM features into our product?

Yes. We build AI features into real products: assistants and copilots, document extraction and classification, search and summarization, and agents that take actions. We work model-agnostic through the Vercel AI Gateway, so your product can use Claude, GPT, or Gemini and switch between them without a rewrite. We handle prompts, tool calling, retrieval, streaming, evaluation, and the guardrails that keep it reliable, and we are honest about where AI helps and where it does not.

AI is a core part of what we build, in our own products and client work. We build model-agnostic through a gateway rather than betting the product on one provider, because this market changes fast.

Two shapes this usually takes

AI work tends to land in one of two shapes. The plumbing is similar, but the design and the risks differ, so we name it early.

AI inside your product

Feature

Intelligence added to an app you have or are building: a copilot, smart search, drafting, extraction, or classification.

Language: TypeScript, via the Vercel AI SDK and AI Gateway
How it ships: Runs in your existing app and backend, streaming responses to the user in real time.

What it gives you

Model-agnostic access to Claude, GPT, and Gemini
Streaming responses
Tool calling into your own functions
Retrieval over your data (RAG)
Structured output you can trust

What you can do with it

Copilots and assistants
Draft generation and rewriting
Search and summarization
Extraction and classification

Pick this when

You have a product and want to make part of it smarter, without betting the whole thing on AI.

Agents that do work

Agent

A system that takes a goal and carries out multi-step work with tools, not just a chat reply.

Language: TypeScript, with a durable execution layer for long jobs
How it ships: Runs as a service with tools, memory, and scheduled or triggered runs.

What it gives you

Tool use and function calling
Memory and state across steps
Human-in-the-loop checkpoints
Scheduled and event-triggered runs
Logging and evaluation

What you can do with it

Support triage and follow-ups
Ops and reporting automation
Research and monitoring
Anything repetitive and rule-based

Pick this when

You want AI to do a job end to end, reliably, with a human in the loop where it matters. This overlaps with our AI Automation service.

What we build with AI

Copilots and in-product assistants
Document extraction and classification
Search, summarization, and Q&A over your data
Agents that take actions with tools
Brand and content monitoring across AI search
AI features wired into an existing web or mobile app

Our approach

We treat models as a component, not the product. That means building model-agnostic through a gateway so we use the best model for each job and swap providers as the market moves, keeping a human in the loop where judgment matters, and measuring output with evaluations rather than vibes. We are also honest about the ceiling: if a task does not actually need AI, or the reliability is not there yet, we will tell you and save you the spend.

How an AI build runs

1
Find the real use
We pin down where AI creates value and where it does not. The cheapest AI is the feature you decide not to build.
2
Prototype the hard part
We build the riskiest prompt or workflow first and put real inputs through it, so we learn the quality early.
3
Wire it in
Gateway, streaming, tool calling, and retrieval, built into your product behind clean interfaces.
4
Evaluate and guardrail
Test sets, guardrails, and fallbacks, so it behaves when inputs get weird and fails safely when it must.
5
Ship and watch
Launch with logging and cost controls, then tune with real usage in front of us.

What to know before you build

AI is powerful and oversold in equal measure. Here is the honest state of it for a product team.

Models make mistakes: Language models are confidently wrong sometimes. We design for it with guardrails, structured output, and human review where the cost of an error is real, rather than pretending it does not happen.
Cost scales with use: Tokens are a real, variable cost. We build with the right-sized model per job and cost controls, so a popular feature does not become a surprise bill.
The market moves monthly: Providers and model rankings change constantly. Building model-agnostic through a gateway is how we keep you from being locked to a bet that ages badly.
Not everything needs AI: Some problems are better solved with plain code. We will tell you when that is the case instead of selling you a model you do not need.

AI & LLM Apps questions, answered

Which AI models do you use?: Whichever fits the job. We work model-agnostic through the Vercel AI Gateway, so we can use Claude, GPT, or Gemini and switch between them without rewriting your app.
Can you add AI to our existing product?: Yes. Most of our AI work is features wired into an app that already exists, behind clean interfaces so it is easy to change later.
How do you keep it from making things up?: Retrieval over your own data, structured output, guardrails, and human review where errors are costly. We also measure quality with evaluations rather than guessing.
What does it cost to run?: It depends on usage and the model. We build with cost controls and right-sized models, and we can pass token costs through transparently or bundle them, whichever you prefer.
Is this the same as your AI Automation service?: Related. This page is about building AI into products. AI Automation is about removing repetitive work in your operations. They use the same tools and often run together.
Do we own what you build?: Completely. The code, the prompts, and the configuration are yours.

Official documentation

Primary sources for the tools we build with.

Vercel AI SDK
The toolkit we build AI features with.
Vercel AI Gateway
Model-agnostic access and failover across providers.
Anthropic Claude documentation
One of the models we build on.
OpenAI documentation
Another of the models we build on.

Related at Foundri

AI & LLM Apps

Adding AI to your product?

Tell us what you are trying to do. We will tell you where AI helps, where it does not, and how we would build it.

Talk about an AI build See our work