A purpose-built college assistant — grounded in your catalog, policies, and website. Answers in seconds, in your voice. No student private data is stored or released.
Built around one job, done well: answering student questions from your catalog, policies, and website — adapts to the browser language, on your own domain. Nothing made up, nothing guessed.
The bot is reviewed by your team, runs every message through a safety pipeline, and bypasses the LLM entirely on crisis keywords to surface 988 (helpline number) or your counseling center in one tap.
Easy questions answer instantly. Hard ones stream as the model reasons.
Add catalog PDFs, policy handbooks, advisor-only memos, even internal FAQs. We chunk them, embed them, and the bot cites the right page back to the student.
The chatbot on your website, using your branding. Students never leave your site — and the conversation never leaves your data perimeter.
WCAG 2.2 AA. Focus trap. Keyboard-first. Screen-reader friendly.
Every conversation is logged for your team to review. You get summaries of what your users are asking. Database is rebuilt nightly so registration dates and tuition reflect today, not two years ago.
Answers only from your catalog, your policies, your current schedule, your faculty directory. When it doesn't know, it says so — and hands off to a human.
Connected to your website and event calendars. Registration windows, financial aid dates, tuition, and course offerings reflect today — not whatever was on the internet two years ago.
The student never leaves your site, and the conversation stays as your data — not training material for someone else's model.
You see what students are asking, where they're stuck, and where your website is failing them. Every conversation is logged, auditable, and reviewed by your team.
A general chatbot is a tourist.
Yours is staff.
Typical pilot-to-production timeline: 30 days. No long-term contract for the pilot.
We bring in your catalog, policies, FAQs, and website content. No scraping — we work with your web team. Upload PDFs and DOCX policy handbooks straight into the index.
sources:
- catalog.your-college.edu
- policies/handbook.pdf
- website://articles
- faqs/registrar.docxPersona, scope, escalation paths, and refusal rules are set with your team. Brand voice. Out-of-scope handoffs to the right advisor. Quiet hours. All reviewable, all editable.
persona: "warm, factual, brief"
scope: ["admissions", "aid", "registrar"]
escalate_to:
finance: aid@your-college.edu
crisis: 988 + counseling centerLaunch on one department or audience — usually admissions or financial aid — for 60 days. We hand you a dashboard of what students actually asked and where your site sent them in circles.
pilot:
department: admissions
duration: 60 days
cost: $0
contract: none requiredScale based on what the pilot data actually shows — never on what we promised. Typical pilot-to-production timeline: 30 days after the pilot ends.
expand:
↳ pilot showed 72% deflection
↳ add: registrar, advising
↳ go-live: 30 days
↳ contract: month-to-monthEnough volume that staff is drowning in repeats.
Information is there — but no one can find it.
Answering the same five questions by phone for the third time today.
Not the right fit? We'll say so. We've turned down deployments where the fit was wrong — your time is better spent than on a tool that won't move the needle.
Every feature came from a real conversation with a real higher-ed team. No hype, no magic — just the pipeline you'd build if you had a year and the right ML hires.
Students wait 3 days for an answer their college's website already has.
Instant, cited answers — pulled live from your own admissions, aid, and registrar pages.
Off-the-shelf chatbots hallucinate tuition, deadlines, and program names.
Hallucination guard verifies every phone number, dollar amount, course code, and URL against retrieved chunks.
Your existing chatbot doesn't speak Spanish, Vietnamese, or Mandarin.
Multilingual by default. Same grounded pipeline, native language for every student.
Procurement says no SaaS for student data.
Run on your hardware. Ollama-compatible LLMs. SQLite, Chroma, BM25 all local. Nothing leaves the campus.
“We went from a 3-day email queue to instant, cited answers. Our staff finally got their afternoons back.”
“Procurement was the hardest sell. On-prem with FERPA-aware logs is what got it across the line.”
“I have never seen a vendor ship a chatbot that actually cites the source paragraph. This is what students deserve.”
“Our Spanish-speaking parents stopped calling the front desk at 7am. That alone justified the pilot.”
We price on conversations, not enrollment. A pilot that succeeds shouldn't cost more than one that fails. Full pricing is shared after we've seen your scope — too many variables to put a single number on the page.
One department or audience — usually admissions or financial aid. We turn it on, you watch what students actually ask.
Full rollout across your domain. Pricing scales with conversations, not enrollment — so an honest pilot doesn't punish a successful pilot.
Your servers, your network, your audit trail. Runs entirely on your hardware — no student data leaves the campus. We handle deployment, retraining, and ops.
Why no number on the page? Public-sector budgets need a real quote, not a sticker. Campus pricing scales with actual usage and the LLM provider you choose (local Ollama, hosted, or your own inference endpoint). We'll send a one-page quote within 48 hours of the pilot kickoff.
Two weeks, on us, on one department or audience of your choosing — usually admissions or financial aid. We provision a hosted demo URL within 48 hours, you point internal users at it for two weeks, and we hand you a digest of what students actually asked. No credit card, no contract, no auto-conversion to paid. If you want to extend or expand into a 60-day production-style pilot, that's a separate conversation.
· Let's talk ·
Book a 30-minute pilot conversation. We'll show you what your students are actually asking — and where your site is sending them in circles.
Honest, no-nonsense privacy: no FERPA-covered, student-identified data is stored in our model or our logs. You see every conversation. You own every word.