AI agents that run on your own systems

Your data and control

• Can we use AI without our data leaving our servers?

  Yes. We deploy agents in your own cloud account (AWS, Azure, GCP) or on your own servers, so data stays where you decide. You also choose the model provider, including open-source models running inside your environment, so nothing goes to a third-party API unless you allow it. If you need hardware for your own servers, we work with a hardware partner.

• Will the AI agent do the task the same way every time?

  Yes. Each agent follows a fixed set of steps in the same order, under the same rules, every time. The AI is used only where judgment is needed, such as reading a message or summarizing a call. Before release, each agent is tested against saved cases, and a release is stopped if quality drops.

• How do you stop an AI agent from exposing sensitive data?

  Each agent reaches only the systems and fields it needs, and your rules are built into the workflow. Examples:

  • masking ID numbers or financial details before they reach a model
  • blocking certain data from leaving a system
  • holding an action until a person approves it

  Every step the agent takes is recorded, so you can see what it touched and when.

• Can we see a record of everything the AI agent did?

  Yes. Every run produces a timeline of each step, with the input it received, the output it produced and its timing. When the agent runs in your own cloud account, that record stays there too, so your compliance team or auditors can review it directly.

• Can the agent write directly into our existing systems?

  Yes. The agent reads from and writes to your systems of record, such as your CRM, practice management software, POS or ERP, so no one re-enters a message by hand. Stackbone comes with hundreds of connectors, and we build custom integrations where none exists. At 4Geeks Academy, the agent writes a tag, score and summary for every lead into ActiveCampaign.

Working with us

• We have our own tech team. Why work with you?

  We build the first agent with your team, then hand over the source code and walk them through it. At 4Geeks Academy, their team shipped a second agent on the same foundation the week after go-live. From there, your team keeps building, with us or on its own.

• Why not use the AI our software vendor already sells?

  Use it where it covers the job. A vendor's built-in AI works with that vendor's data. Our agents work across all of your systems, so one workflow can read your CRM, write to your scheduling tool and report from both. You also choose the AI provider, and the code is yours.

• Can we export our agents if we stop working with you?

  Yes. Your agents are code in a repository you own, and we hand over the source code with a walkthrough of how it works. If you stop working with us, the code stays with you.

• How much does a custom AI agent cost?

  Most single-agent builds cost between $5,000 and $20,000. The price depends on how many systems the agent connects to and how much judgment the task needs. Running costs start at $200 a month and depend on how much work the agent does, mainly AI usage per conversation or document. We estimate them before you sign. We scope the first agent on a free call and give a fixed quote before any work starts.

• Do you just build the agent, or do you train our team?

  Both. We build the first agent with you and keep it running. After that, you choose: we build the next ones, or we train your team to build them on the same platform.

• Is Stackbone a wrapper on n8n or Zapier?

  No. Stackbone is its own platform: a TypeScript SDK and CLI for building agents and workflows, and a managed runtime that runs them. Each step runs once, saves its result and retries on failure. A run survives crashes and redeploys, and can wait days for an approval. Technical details are at docs.stackbone.ai.

Education

• Can an AI voice agent handle admissions calls?

  Yes. A voice agent can:

  • answer and place admissions calls
  • answer program questions from your own course information
  • qualify the applicant and book a call with an advisor
  • write the outcome into your CRM

  Every layer (speech-to-text, LLM, text-to-speech) can be swapped, so you can move to a more natural voice or a different model without rebuilding. We built The Big School's admissions voice agent this way, so they can change voice or model providers without starting over.

• Can AI answer parent inquiries instantly and book a trial class?

  Yes. The agent replies over WhatsApp, SMS or email at any hour. It:

  • collects the student's age, level and goals
  • answers questions from your program information
  • books a trial class into your calendar
  • passes the conversation to a person when a family wants one

  4Geeks Academy gets about 30–50 WhatsApp contacts a day. Their agent replies to each one and writes the lead into ActiveCampaign.

• Will an AI agent replace our admissions staff?

  No. It takes over the admin: the first response, qualification and scheduling. That leaves your team's time for applicants who are ready to enroll. You decide which steps stay human. For example, the agent qualifies and books, and an advisor holds the enrollment conversation. At 4Geeks Academy, two full-time staff moved to higher-value work.

• Is an AI admissions or tutoring agent FERPA compliant?

  The setup keeps student records under your school's direct control. The agent runs in your own cloud account, reaches only the records you allow, uses only the model providers you approve, and records every step. Your school stays responsible for how FERPA applies, and the setup lets you show that control.

Financial services & investment

• Can AI analyze a data room like an analyst would?

  Yes, as a first pass that your team reviews. An agent can:

  • read every document in the data room
  • extract the key figures
  • cross-check claims against the source documents and the web
  • flag inconsistencies, with citations back to the page

  We built Solon, our own meeting agent, to do this live: it joins investment meetings with the data room loaded and checks claims as they're made.

• How do we keep deal documents confidential when using AI?

  Run the AI where the documents already live. We deploy in your own cloud account or on your own servers, using model providers you approve or open-source models inside your environment. Each agent reaches only the systems you allow, and every step it takes is recorded.

• What records does an AI agent in client meetings create under SEC and FINRA rules?

  Transcripts, summaries and any messages the agent drafts may count as business records under SEC Rule 204-2 and FINRA recordkeeping rules. When the agent runs in your own environment, those outputs and the step-by-step record of each run stay there. Your compliance team applies its own retention and archiving, and decides what is kept and for how long.

• What workflows can a financial advisory firm automate with AI agents?

  Common starting points:

  • client onboarding and document collection
  • meeting preparation from the CRM and portfolio data
  • meeting notes written back to the CRM
  • follow-up drafting for advisor approval
  • recurring client reporting

  Each one follows the same steps for every client, and every step is recorded.

Manufacturing & industrial

• How do you capture tribal knowledge before an expert retires?

  Capture it from the work they already do. An agent can:

  • ingest the expert's calls, email and messages with customers and suppliers
  • extract the decisions, specifications and rules of thumb
  • organize them into a searchable knowledge base

  Your team can ask that knowledge base questions and get answers with the source attached. Future agents can build on it.

• I bought a business and the previous owner leaves in six months. How do we keep what they know?

  Start capturing now, while they're still taking calls. For a manufacturer that acquired a specialty parts maker, we built a system that:

  • routes the previous owner's customer communications through new company channels
  • extracts the know-how from those conversations
  • documents it in a structured knowledge base the new team uses

  The same approach can cover phone calls, email, SMS, WhatsApp and Telegram.

• How do we train an AI on everything our company knows?

  Connect it to where the knowledge already lives, and structure it. We connect to your documents, ERP, email and call records, extract what matters, and organize it into a knowledge base with sources attached to every answer. The knowledge base can stay in your own environment, and later agents can draw on it.

Restaurant chains

• Can AI combine reporting across all our locations and systems?

  Yes. An agent pulls data from your POS, labor, inventory and delivery platforms across every location, reconciles it, and produces the reports your operators need. It also answers questions such as "which stores missed labor targets last week?" with the source data attached.

• Can AI take phone orders straight into our POS?

  Yes. We can build a voice agent that answers calls during rush hours, takes the order, confirms it with the customer and sends it into your POS (Toast, Clover or another) through a connector or a custom integration. Delicate calls, such as a complaint, an allergy question or a large catering order, go to a staff member with the order details and a call summary already in hand.

Dental & aesthetic clinics

• Can an AI receptionist book directly into Dentrix or Open Dental?

  Yes. Through a connector or a custom integration, we can build the agent to:

  • answer calls and messages
  • check real availability in your practice management system
  • book, confirm and reschedule appointments there directly

  It writes to the schedule, so the front desk doesn't have to re-enter a message. After hours, it handles new-patient calls that would otherwise go to voicemail.

• Is an AI receptionist HIPAA compliant?

  The setup keeps patient data under your control. The agent runs in your own environment, reaches only what each step needs, hides sensitive details before they reach a model, and records every step. Where a third-party model is used, we can choose providers that offer a business associate agreement (BAA). Or we run the model inside your environment.

• Will patients be annoyed by an AI receptionist?

  Not if it answers quickly and hands off well. You choose the most natural voice available and can switch it later. The agent transfers to staff as soon as a patient asks for a person or a request falls outside its rules.

Professional services

• How can law firms use AI without breaking client confidentiality?

  Keep client data inside the firm's control. We deploy agents in your own environment, use only model providers you approve, limit each agent to the systems it needs, and record every step it takes. ABA Formal Opinion 512 (American Bar Association, 29 July 2024) expects lawyers to understand how AI tools handle client information, and this setup lets you show where that information goes. Consent and supervision remain the firm's decision.

• Can AI client intake integrate with Clio?

  Yes. We can build an intake agent that:

  • answers inquiries by phone, chat or email
  • flags possible conflicts against your existing contacts for an attorney to confirm
  • collects the matter details
  • creates the contact and matter in Clio (or your practice management system)

  Qualified leads are booked with an attorney, and every intake is recorded with the source conversation.

• What can accounting firms automate with AI agents?

  Common starting points:

  • client document collection and follow-up
  • matching transactions across bank, ledger and invoice systems
  • month-end close checklists
  • drafting client reports for review

  Each workflow follows the same steps every time and records them, so your auditors can review what the AI did.