Give your office a second brain that runs on your own server
It remembers every file, email and meeting your team has had, sorts the paperwork as it arrives, and answers you in the chat app you already use. We pick the hardware, install the models and agents, connect your apps and set up access from your phone. On-site across the Sacramento area, remote anywhere in the US.
What does local AI setup cost in Sacramento?
Setting up one machine costs $499, with the on-site visit included anywhere in our service area. A server for a team starts from $1,000. The hardware check before either is free. Care after setup is $300 a month. Hardware, cloud subscriptions and fees for outside AI models are separate: you buy them in your own name, and we tell you exactly what to buy before you spend anything.
Pick your first superpower
The $499 setup includes one of these, chosen with you after the free check. Each one after that is added the same way, and they all share the same machine.
Second Brain
Everything your team has written in one place it can read: files, notes, emails, chat history. Ask "what did we agree with the landlord about parking?" and get the paragraph and the file it came from. Built for a whole company with permissions and many sources, it becomes its own project: private AI on your documents.
Document Organizer
Scan a letter or drop a PDF into a folder. It gets read, renamed, tagged, filed in the right place and becomes searchable by what it says. No more "Scan_0047.pdf".
Pocket Assistant
An assistant that remembers everything you told it, in WhatsApp, Telegram, Slack, Signal or iMessage. It reads your inbox, checks the calendar, drafts replies, books the meeting and reminds you about the thing you forgot.
Morning Briefing
At 7am, one message: today's meetings, emails still waiting for an answer, invoices due this week, anything that changed overnight.
Night Shift
Jobs that run while you sleep: overnight it sorts the day's documents, checks what came in and has the summary ready before you open the laptop.
Meeting Memory
Recordings transcribed on your own machine, with the summary and action items in your inbox before you are back at your desk. Recorded with everyone's consent, as California law requires.
Private AI Chat
A ChatGPT-style chat window for the whole team, a login for each person, answers from a model running on your machine. Paste the client file without a second thought: nothing typed into it leaves the building.
Workflow Autopilot
A web form becomes a CRM record and a calendar slot. An invoice PDF becomes a row in your books. A new order becomes a summary in Slack. Nobody retypes anything.
Talking Office
"Is the back door locked?" "Turn off the lights in the conference room." Local voice control through Home Assistant, no cloud speaker listening. More in devices and smart spaces.
Night Watch
Cameras that know what they see: an alert for a person at the back door after hours, silence for every passing car. Detection runs on your hardware, with no camera subscription.
Code Partner
A code assistant for your developers. Continue, Tabby or Aider with a model on the office server keep your source code in-house; Claude Code and Codex work through your own cloud keys.
Image Studio
Product shots, mockups and graphics generated on your own graphics card, as many as you want, with no per-image fees.
By line of work
Case Timeline
Deposition transcripts and discovery files turned into a dated timeline, with the places where witnesses' accounts differ marked for review.
Contract Checker
Incoming contracts compared with your standard terms, every clause that differs listed for an attorney to check.
Tax Season Tracker
A checklist for each client: which W-2, 1099 and 1098 forms arrived, whether each scan is readable, what is still missing, with a drafted request you approve before it goes out.
Statement Matcher
Bank statement lines matched to the books, with a short list of questions about the charges nobody recognizes.
Repair Desk
A tenant sends a photo of the leak. The request gets an urgency level, a drafted work order for your vendor and an updated ticket.
Lease Digest
Leases boiled down to one table: rent, escalations, renewal options, CAM charges and notice dates.
Plan Reader
Plan sets and spec PDFs read page by page into a scope summary and quantities per trade, ready for bid requests to subs.
Call Notes
Recordings from your phone system transcribed on your machine, with notes and tasks added to the CRM. Recorded only with everyone's consent.
Privacy Filter
Social Security numbers, account numbers and addresses masked in PDFs before a file goes to an outside party.
Agents and platforms we install
Agents that read email and use tools work best with a cloud model or a large local one. Small local models handle chat, summaries and search. The written check tells you which your hardware can carry.
| Agent | What it is good at |
|---|---|
| OpenClaw | The best-known personal AI agent. Lives in your chat apps, works with email, calendar, files and the browser, runs jobs on a schedule. |
| Hermes Agent (Nous Research) | Remembers past conversations and writes its own skills as it learns your work. Reaches you on Telegram, Slack, WhatsApp, Signal, Microsoft Teams and 20+ platforms. Works with local or cloud models. |
| NanoClaw | A lighter agent built on Claude, where every agent runs in its own container. A good choice when the agent reads email from strangers. |
| ZeroClaw, PicoClaw | Small, fast agents for modest hardware, down to low-power boards. |
| Cardloop | Our own open-source agent cockpit (MIT): a kanban board where a card starts an agent, with a web and Android app. GitHub |
| n8n | Visual workflow automation with AI agent steps and hundreds of ready integrations. Self-hosted, so the data stays with you. |
| Dify, Flowise, Langflow | Platforms, not ready agents: build your own assistants and internal AI apps by connecting blocks. We build the first one with you. |
| AnythingLLM, Open WebUI, LibreChat | Private chat for the team, with document upload and per-user accounts. |
| Continue, Tabby, Aider | Code assistants in your developers' editors, running on local models. |
| Claude Code, Codex CLI | Coding agents on your server, working through your own Anthropic or OpenAI key. |
Not sure which one fits? That is what the free check is for. Most setups combine two: a chat for the team and one agent for the person who needs it most.
Models we run
- Language models: Llama, Qwen, Gemma, Mistral, DeepSeek, gpt-oss, Phi, in sizes matched to your hardware.
- Speech: Whisper for transcription, Piper for natural voice output.
- Vision: models that read scans, screenshots, photos and camera frames.
- Images: Stable Diffusion and Flux workflows in ComfyUI.
- Cloud models when you want them: Claude, GPT and Gemini through your own API key, for tasks a local model can't handle well.
Runtimes: Ollama, llama.cpp, vLLM, LM Studio, MLX on Apple Silicon. We choose the one that gets the most out of your machine.
What it connects to
- Chat: Telegram, Slack, Microsoft Teams, Discord, Google Chat, Matrix; WhatsApp and Signal on a separate number; iMessage through an always-on Mac
- Email and calendar: Gmail and Google Calendar, Outlook and Microsoft 365, any IMAP mailbox
- Files: Google Drive, OneDrive and SharePoint, Dropbox, network shares, Synology and QNAP NAS
- Business apps: QuickBooks, HubSpot, Salesforce, Zoho, Airtable, Notion, Google Sheets, Asana, Trello, Jira, GitHub
- Databases: PostgreSQL, MySQL, SQL Server, SQLite
- Your phones: AI that answers calls and books appointments, see AI for calls and chat
- Industry software: Clio, Smokeball, NetDocuments, Karbon, Xero, Gusto, Buildium, DocuSign, Dotloop, Procore, RingCentral, Dialpad, Zoom Phone, 3CX. Through their APIs, or through exports and shared folders where a product has no API
- Anything with an API: through MCP servers (the open standard agents use to reach tools) or a custom connector we write
Hardware: what you have, or what we spec
We work with any server. If you already own the machine, we start there.
- Apple: Mac mini, Mac Studio, MacBook Pro. Quiet, efficient, and their shared memory runs large models.
- PCs and workstations with NVIDIA graphics, including the gaming PC in the corner. The GPU does the heavy lifting.
- Compact AI desktops: NVIDIA DGX Spark, mini PCs on AMD Ryzen AI Max with up to 128 GB of memory.
- Tower and rack servers: Dell PowerEdge, HPE ProLiant, Lenovo ThinkSystem, Supermicro.
- Servers you already run: Proxmox, VMware, Hyper-V, Unraid, TrueNAS, Windows Server, any Linux.
- NAS boxes: Synology and QNAP for storage, document pipelines and lighter agents.
- Your cloud: AWS, Azure, Google Cloud, Hetzner, DigitalOcean, or a GPU cloud, in your account.
Buying new? We write a hardware list for your budget with today's prices, you order it in your name, and the warranty is yours.
Connect the servers you already own
A GPU machine in one room, a NAS in the closet, a cloud server from three years ago: we tie them into one system on a private network.
- Models on the GPU machine, agents on a small always-on box, files on the NAS, all working as one
- The office server and your home office machine on the same private network, so you reach both from anywhere
- One AI server shared by two locations
- An existing cloud server added to the same network, so the agent can work with what runs there
Your phone and laptop join that network with a small app. Nothing gets opened to the internet.
Where your data goes
Where your words and files are processed, for each kind of setup.
Fully local
On your machine. Works with the internet unplugged.
Cloud model, your key
Prompts go to the model provider you pick (Anthropic, OpenAI, Google), under your account and their terms.
Mixed
Routine work stays local; only the tasks you choose go to a cloud model.
The free written check says which one your tasks need, before you buy anything.
How a setup goes
Free hardware check, by email
Send photos and specs of what you have, or tell us what you want to buy. We reply by email with a written answer: what your hardware can run, what to buy if anything, where your data will go, how many people can use it at once, a fixed price and a date.
Setup
We come to you, or we configure the machine in our office on Watt Ave and bring it over, or we set it up remotely on your server or cloud.
Acceptance tests
Your real examples run twice. The machine is restarted and everything comes back on its own. A backup of your settings and data is restored and the assistant works from it. On team setups, we check that one person can't see another's files or chats.
Handover
Every login in your name, a written runbook, and a walkthrough for the people who will use it. Our temporary access is removed unless you choose Care.
Set up so it stays private
- No port forwarding. Remote access runs through a private network (Tailscale or WireGuard), the same way for every device.
- Agents run in their own container or virtual machine and can reach only the folders, apps and addresses we list together.
- Sending email, paying, deleting or publishing needs your confirmation. We test that the agent can't get around it through another tool.
- Only vetted skills and plugins, with versions written down. Nothing installed from a public catalog on a whim.
- Encrypted backups of your settings, chats and data, stored in a separate place.
- Every account, key and password is yours.
Packages
Free hardware check
A written review by email of what you have and what you want, with a fixed quote.
One machine
One computer or server, private chat with local models, one assistant or agent for one scenario, connected to up to two of your apps (for example Gmail and Google Calendar), private-network access for up to 5 devices, backups, one on-site visit, walkthrough.
Team server
Shared server for a team, separate accounts with tested separation, up to two assistants or agents, monitoring.
Your server or cloud
Either package on hardware or cloud you already run, set up remotely anywhere in the US.
Care
Model and software updates, backup checks, monitoring, small changes, done remotely.
Private AI on your documents
A full system that answers from your company's documents is a separate project: private AI on your documents.
Not in the price: hardware, cloud subscriptions, outside AI model fees. Virtualization with a graphics card passed through, and anything beyond the package, is quoted in writing after the check. The full company price list is on the pricing page.
For an office or a home office
Law practices, accountants, real estate teams, engineers and owners who work from home all fit. If the work involves client files, contracts or anything you would not paste into a public chatbot, a machine you own is the simplest answer. We sign the agreement online before the visit.
What we don't do
- Cabling inside walls or mounting. A licensed low-voltage contractor does that under your own contract; we connect to the network you have.
- Resell hardware or build PCs from parts. You buy finished machines in your name.
- Keep remote access to machines that hold patient records. For clinics we install and test with sample data, and your administrator holds every login.
- Promise that every answer a model gives is correct. We stand behind the system passing the tests we agreed on; you review what it writes before it goes out.
Service area
On-site: Sacramento, Carmichael, Citrus Heights, Rancho Cordova, Arden-Arcade, Roseville, Folsom, Elk Grove and Davis. Remote setup anywhere in the US. Our office is at 3550 Watt Ave, Suite 408, Sacramento.
Frequently asked questions
What people ask before we set up AI on their own hardware.
Do I need a graphics card?
Not always. A Mac with enough memory or a recent mini PC runs smaller models well. For bigger models, several people at once, or images, a GPU or a larger Mac makes the difference. The free check tells you which.
Can you use the computer we already have?
Usually yes. Send the specs and we tell you what it can run. If it falls short, you get a list of what to add or buy.
Will it be as good as ChatGPT?
For drafting, summarizing, searching your files and routine questions, a well-chosen local model on decent hardware handles everyday office work. For the hardest tasks we connect a cloud model with your own key, and you decide which tasks may use it.
Does my data leave the building?
With a fully local setup, no. With a cloud model, only the prompts you route to it. The written quote says which setup you have.
Can you set up OpenClaw or Hermes Agent?
Yes, along with NanoClaw, n8n and the others above. We install them in an isolated container, connect only what you approve, and turn on confirmation for anything that sends, pays or deletes.
Can I reach it from my phone outside the office?
Yes. Your phone joins the private network with a small app, and the assistant also answers in the chat app you already use.
How much does the hardware cost?
It depends on the models and how many people use them. After the free check you get a list with current prices for your budget, before you buy anything.
Can it sit in my office? What about noise and heat?
Mac and mini PC setups are quiet enough for a desk. GPU towers are louder and warmer, and the written check says where one should go.
Who owns the hardware and the logins?
You do. You buy the machine in your name, and every account, key and password is handed over at the end.
Do you work outside Sacramento?
On-site visits cover the Sacramento area. Anywhere else in the US we set up remotely on your server or cloud, or configure the machine here and ship it to you at your cost.
Who sets it up
Our team runs its own self-hosted infrastructure: several Proxmox hosts behind Cloudflare and Traefik, one registry of services, weekly integrity checks and offsite backups.
Cardloop, our own open-source agent cockpit (MIT), is on GitHub: a kanban board where a card starts an agent, with a web and Android app.
Ready to pick your first superpower?
Tell us what you have and what eats your week. The check is free and comes back by email as a written answer: what your hardware can run, what to buy if anything, and a fixed price.
Get my free hardware checkOr call (916) 618-0502.
Tell us what's slowing you down.
We reply the same business day.