Telegram front door · Five platform layers

More than a chatbot — a full student stack.

Emily combines a learning engine, campus layer, app integrations, and student-controlled memory — behind enforced academic boundaries. Telegram is how you reach it, not what it is.

Open in Telegram Platform features Trust & safety

Beta: Emily is in active development — features may change and bugs can happen. Report issues or email support@emily-ai.net.

Platform layers: 5
Study workflows: 3+
OAuth integrations: 4
Operating modes: 3

See it in action

One Telegram thread

Voice notes in, grounded replies out — study help, proofreading, and campus workflows without leaving chat. Emily runs on dedicated infrastructure; Telegram is the front door.

· Text & voice messages
· Files & lecture audio
· Live location share
· Scheduled briefings

Open in Telegram

Emily AI

@Emcbearbot · online · voice on

Primary

Message Emily…

Why Emily instead of ChatGPT?

Purpose-built for university students — the same LLM stack is not the same product when guardrails, campus workflows, and mastery tracking are enforced end to end.

Generic ChatGPT

Emily AI

Generic ChatGPT
General-purpose chat
Emily AI
Academic + campus scopes with enforced guardrails
General-purpose chat
Academic + campus scopes with enforced guardrails
Generic ChatGPT
One-off answers
Emily AI
Learning + quiz sessions with mastery and spaced review
One-off answers
Learning + quiz sessions with mastery and spaced review
Generic ChatGPT
No multimodal study flow
Emily AI
PDF ingest → grounded summary → voice reply → quiz → Spotify
No multimodal study flow
PDF ingest → grounded summary → voice reply → quiz → Spotify
Generic ChatGPT
No campus tooling
Emily AI
Timetable, wellness briefings, health KB, geofence, morning routines
No campus tooling
Timetable, wellness briefings, health KB, geofence, morning routines
Generic ChatGPT
No app connections
Emily AI
Google Mail/Calendar/Drive, Notion, Spotify, Uber
No app connections
Google Mail/Calendar/Drive, Notion, Spotify, Uber
Generic ChatGPT
Opaque memory
Emily AI
Persistent memory you list, export, and forget across sessions
Opaque memory
Persistent memory you list, export, and forget across sessions
Generic ChatGPT
Provider-default safety
Emily AI
Local logs, parent escalation, student appeals
Provider-default safety
Local logs, parent escalation, student appeals
Generic ChatGPT
Answers may hallucinate
Emily AI
Primary grounding stack + upload citations
Answers may hallucinate
Primary grounding stack + upload citations

Not a tutor prompt on top of a general model — enforced modes, memory controls, and campus routing. FAQ →

What's under the hood

Five platform layers stack beneath one Telegram thread. Select a layer to see what Emily actually ships — beyond generic Q&A.

Access channel

Telegram — text, voice, files, location

↓

Learning engine

Mastery, not answer dumps

Structured Primary, Learning, and quiz flows with SQLite mastery, spaced review nudges, proofreading rubrics, and grounded file ingest.

Learning sessions + quiz me on that bridge
Proofreading scores & numbered fixes (no ghostwriting)
Uploads & voice → grounded Primary replies
Illustrations & science subject coverage

Explore learning engine →

Persistent memory & chat limits

Persistent memory survives across Telegram sessions until you delete it. The rolling chat window and token budget only bound how much raw conversation is replayed each turn.

Persistent memory

Remembers you across sessions — not just this chat

Emily keeps a lightweight persistent student model on the tutor server: preferences, professors, pinned course notes, and mastery progress. It reloads when you open Telegram again — not a one-off ChatGPT thread that resets when you close the app.

Cross-session: Yes
KB sources: 5 max
Saved facts: ~12
Until: You forget

What persists	Examples	Until
Preferences & facts	K-pop, study spots, habits	Until /forget
People & professor prefs	Dr. Kim, APA formatting rules	Until /forget
Pinned kb: sources	Syllabus, Drive/Notion notes	Until removed
Mastery & profile	Quiz scores, review dates, USER.md	Exportable · forgettable

Survives across sessions. Saved on the tutor server — close Telegram, come back tomorrow, or survive a server restart. Persistent until you list and delete it.
Tell Emily explicitly. remember I like … · pin chem syllabus as kb: … · remember Dr. Kim is my chem professor — written immediately when privacy mode is off.
Passive learning (with consent). Mention a preference in chat — Emily may ask yes/no before saving. High-confidence facts can auto-save with a short note.
You stay in control. what do you know about me? · /memory · /forget everything · /exportdata — list, export, or wipe persistent memory anytime.

Try: remember I like K-pop · remember I usually study at Weldon · remember Dr. Kim is my chem professor

Persistent memory is not the rolling chat log. Facts, kb: sources, and mastery stay on the tutor server until you delete them. The turn window (24h TTL, token budget) only bounds how much raw conversation is replayed each reply — trimming chat does not erase what you explicitly saved.

Full persistent memory catalog → · Chat window limits →

Token management

Built-in limits — not infinite chat

Emily is affordability-first by design. Context, reply length, session history, and auxiliary model calls are budgeted so students get focused tutoring without runaway token burn or endless log retention.

Chat context budget: ~6k
Max reply length: 512
Session TTL: 24h
Daily LLM budgets: Capped

Context budget. Primary-mode prompts fit inside a ~6,000-token budget. If memory, uploads, and history overflow, Emily trims lowest-priority sections first — not a silent infinite context window.
Bounded sessions. Conversation turns are capped (default 12, up to 40 on the tutor server) with a 24-hour TTL. Long threads compress into a session spine summary instead of sending the full log every time.
Capped replies & sidecar calls. Chat replies max out around 512 tokens. Classifiers, mode routing, and passive memory extraction use smaller caps plus daily USD budgets — with a cost breaker when limits are hit.

Trim order when over budget: Mastery snapshot → Session spine → Upload excerpts → Grounding blocks → Older turns — persistent memory and profile kept when possible.

Safety & limits detail → · Persistent memory catalog → · Data rights →

Workflows, not one-off prompts

Multi-step study nights chain ingest, grounding, voice, quiz, and Spotify — mastery tracked the whole way.

Full catalog →

Hint-led practice

Start: /learn · explain then practice · don't give me the answer

Enter — Send /learn or ask in Primary — Emily suggests switching when you want practice.
Classify — Your message maps to a skill (e.g. stoichiometry, mitosis).
Hint + check — Skill-specific hint, then one comprehension question — graded, not answer keys.
Mastery — Scores update; Telegram nudges you when a skill is due for review.

Example: How do I balance this equation? → hint → limiting reagent check → mastery updated.

Built for real stakeholders

Students

After-hours tutoring with hints and quizzes, proofreading, campus directions, and connected apps — without companion-chat drift.

For students →

Parents

Published policies, export and delete in Telegram, guardrails against romantic and explicit use — not surveillance of every message.

For parents →

Schools

Boundary-controlled tutoring layer, auditability, and pilot-ready architecture — teacher augmentation, not replacement.

For schools →

Pilot: Western University sciences students. Testimonials and school logos — add when available.

A platform parents and schools can explain.

Published Privacy Policy and Terms. Students confirm they are 18+ (or parent-managed) before chat.

Contact FAQ All features