Imagine a smarter, quicker version of yourself that the whole company can access, in the field or in the office, 24 hours a day.
A Stone Ridge AI assistant will learn your procedures, your products, your codes, and your jobs. Plus, it can get actual work done in JobTread just as easily as it dispenses advice.
Right now the answers live in your head, a stack of PDFs, and old client threads. Every question that can't find them turns into an interruption for you.
You don't need a custom-trained model. You need one of the best models on the market, connected to a library of everything Stone Ridge Decking knows.
| Custom-trained model | Top model + your playbook | |
|---|---|---|
| Typical cost | From $60,000, plus GPU hosting of $500 to $1,500 a month | $14,500, plus AI credits as you use them |
| Time to build | 5 to 6 months | About 13 weeks |
| When your info changes | Retrain the model: weeks of work and more cost | Add the document and it knows it within minutes |
| Shows its sources | ✕ No | ✓ Yes, the document and page |
| Difficulty | High: data scientists, GPU servers, ongoing retraining | Low: built on a proven platform |
Your playbook is a living library of everything Stone Ridge knows, plus instructions for how you like things done. The assistant reads the right pages before every answer, so it stays current the moment you add something new.
For building code and permit questions, the assistant quotes the exact section and jurisdiction it's relying on, and flags anything your crew should confirm with the county before they build. Your team gets speed without guessing.
Three pieces work together. You feed the playbook, your team talks to the assistant, and the assistant handles the busywork in the tools you already use.
Procedures, catalogs, install guides, code reports, building codes by jurisdiction, and past client conversations. Organized, searchable, and kept current.
With the name, look, and greeting you choose. Staff sign in from a phone or computer and ask their questions in plain English, the way they'd ask you.
It looks up a job, adds a task or daily log, and drafts an estimate from your catalog. It always asks before it saves anything.
Your assistant will run on Proxy's own AI platform, the same engine behind our new assistant in our client portal. That's why this costs a fraction of a from-scratch build, and why it keeps getting better after launch.
While your assistant is hosted with us, every improvement we make to the platform reaches it automatically: new models, better search, new connections, and other timely improvements.
Self-hosting means everything the portal already does has to be set up again as a standalone back end that runs without us.
| What it takes | Standalone setup | With Proxy Portal |
|---|---|---|
| Secure logins and user management (invites, office and field roles) | $2,500 | ✓ Included |
| Gated file library to upload, organize, and search your files | $2,500 | ✓ Included |
| Knowledge bases, staff-only and public, with an editor | $3,000 | ✓ Included |
| Admin dashboard and usage tracking | $1,500 | ✓ Included |
| Hosting setup, backups, and security hardening | $2,500 | ✓ Included |
| Total | $12,000 | Included |
We aim for a usable MVP as soon as possible for testing, enhancing, and adapting as we anticipate future use cases and pitfalls.
| Step | What happens | When |
|---|---|---|
| 1. Plan | Kickoff, goals, and a map of every document, workflow, and question the assistant should handle | Week 1 |
| 2. Gather | Your docs, catalogs, codes, and past client conversations collected and organized into your playbook; JobTread access connected | Weeks 1-2 |
| 3. Design & brand | A simple, field-friendly interface, plus your assistant's name, look, and greeting | Weeks 3-4 |
| 4. MVP | A working assistant: staff logins, your playbook loaded, and answers with sources | Weeks 4-6 |
| 5. Connect | JobTread, the Claude API, the gated file library, and both knowledge bases wired in | Weeks 7-8 |
| 6. Test | Two crew members use it on real jobs, and we log and fix every miss | Weeks 9-10 |
| 7. Fine-tune | Answers, tone, and actions tuned from real use | Weeks 11-12 |
| 8. Launch | Your whole team invited and live | Week 13 |
| Ongoing | Add docs any time, monthly usage check, platform upgrades roll in automatically | Every month |
The phone app installs straight from the browser, with your icon on the home screen. An App Store version can come later if you want one.
We're building you an app: a front end your team uses every day, and a playbook that connects your company's knowledge to the Claude API, all managed through Proxy Portal.
A website chatbot with a help-center knowledge base answers questions. Your assistant also knows your crew, your private procedures, and your jobs.
| Feature | Chatbot + knowledge base | Stone Ridge AI |
|---|---|---|
| Answers questions from a knowledge base | ✓ | ✓ |
| Finds building codes stored in your knowledge base | ✓ | ✓ |
| Public answers for customers on your website | ✓ | ✓ |
| Staff logins with office and field access | ✕ | ✓ |
| Private, staff-only procedures and pricing | ✕ | ✓ |
| Learns from past client conversations | ✕ | ✓ |
| Does real work in JobTread | ✕ | ✓ |
| Installable phone app for the crew | ✕ | ✓ |
The self-hosted and custom-model builds use the same 40 / 30 / 30 schedule.
Once it's live, your team's usage runs on Proxy AI credits, the same credits every Proxy client uses. You only pay for what your team actually asks.
For a team your size, we expect regular use to run about 50 to 150 credits a month, which fits inside the credits your Proxy plan already includes most months, with an optional $50 monthly pack (+100 credits) for busy ones.
Plain English, no fine-print surprises. Here's how ownership works on this project.
Proxy keeps the engine underneath, which is why it keeps improving. Other businesses get their own builds on the same engine. None of them ever see your content.
Once the build is paid in full, Stone Ridge gets a perpetual, royalty-free license to use its assistant for its own business. For this project, that license replaces the standard "while you're a client" platform terms in Proxy's Terms of Service.
If Proxy ever stops running the platform, you get your data and a working copy of your assistant you can run yourself.
It's your app. Start on Proxy Portal and move to your own servers whenever you like by paying the difference between the two builds ($12,000) for the standalone setup. Self-hosted copies are for Stone Ridge's internal use only, with no resale or sharing. You cover your own hosting and AI costs. Updates and support continue only while you're on a Proxy plan.
Sign and we'll start planning this month. Your team will be testing a working MVP within six weeks.