Structured observations
Every note becomes a dated observation per child, tagged by domain and linked to the public milestone reference for their age.
For the community Open Milestones · we're building it now
A free milestone guide for every parent, in their own language.
We're building the world's open milestone guide: every parent, in every language, gets the same quality guidance, with an age-by-age checklist grounded in CDC and WHO guidance and a Claude assistant for their questions and next week's play.
Year-one targets: 100,000 families · 20 languages · zero ads, ever.
- Free forever for individuals
- Open source at launch
- Powered by Claude agents
Not a diagnosis and not medical advice. If a milestone is missed, it encourages you to talk with your pediatrician. It never ranks or labels children.
18-month check-in
Sample- Walking on their own, without holding onMovement
- Pointing to share something interestingSocial
- Trying a few words beyond "mama" and "dada"Language
Mi hija tiene 18 meses y casi no dice palabras. ¿Es normal?
Muchos niños de esta edad dicen pocas palabras y entienden muchas más. Esta semana pueden nombrar juntos lo que ven a la hora del baño o de la comida. Como a esta edad la mayoría intenta decir algunas palabras, vale la pena comentarlo con su pediatra: no para preocuparse, sino para revisarlo juntos.
In English: a parent asks, in Spanish, whether few words at 18 months is normal. The assistant answers in Spanish, suggests naming things together at bath and meal times, and recommends mentioning it to their pediatrician.
What we're building for educators
The Classroom Observation Agent
Teachers notice a lot and have no time to write it down properly. The agent turns quick daily notes and photos into structured developmental observations for every child, drafts weekly updates in each family's language, and flags what a teacher may want to look at more closely. The teacher reviews everything before anything is sent.
Sunflower Room (2–3 yrs) · fictional classroom
PreviewObservation agent
Sample output with fictional children and notes. Click a note to see what the agent would draft.
Weekly parent updates
A warm, plain-language draft for each family, in the family's own language, with one activity to try at home.
Screening flags, for the teacher
When a pattern across weeks may merit a professional screening, the teacher sees the evidence. Never a label, never automatic.
Meeting summaries
Before a parent-teacher meeting, a one-page summary of the term's observations, ready to discuss.
Our goal: a 20-classroom pilot cohort in year one, then every childcare center that wants it. It's a paid product; Open Milestones stays free.
The agent behind it
From a teacher's note to an approved parent update
A multi-step Claude pipeline that runs every school day for every child. Each step calls tools, writes structured output, and hands off to the next. A teacher closes the loop.
-
01Trigger
Teacher saves the day's notes
Typed notes, voice memos or photos from the classroom tablet, or a daily 4 pm batch.
-
02Haiku 5.5
Split, tag, de-identify
Splits group notes per child, swaps names for pseudonymous IDs, tags domains, and turns photos into short text, then discards them.
split_by_childpseudonymizetag_domains -
03Sonnet 5.5
Write the observation
Reads the child's last 8 weeks, matches the age-based milestone reference, and writes a schema-validated observation with evidence and uncertainty.
get_child_historylookup_milestonessave_observation -
04Sonnet 5.5
Draft the parent update
Once a week, one plain-language update per family, written in the family's language, with an at-home activity idea.
get_family_languagedraft_update -
05Haiku 5.5
Safety audit of every draft
Checks for diagnostic wording and comparisons between children, and back-translates to catch drift. A failed draft goes back to step 04.
check_language_policyback_translate -
06Sonnet 5.5
Look for patterns
Across weeks, notices when something may merit a professional screening conversation and prepares the evidence for the teacher.
flag_for_teacher_reviewsummarize_term -
07Human review
The teacher reviews, edits and approves
Every update, flag and meeting summary waits in the teacher's queue. Nothing reaches a family automatically, and screening flags are never shared without the teacher's decision.
queue_for_reviewsend_update(only after approval)
Claude Sonnet 5.5 · planner, writer, reviewer
Structures observations against the child's history, writes parent updates and meeting summaries, answers parents' questions in Open Milestones, and does the multi-week pattern review.
Claude Haiku 5.5 · high-volume triage & auditing
Splits and tags every incoming note, de-identifies it, and audits every draft and every answer for unsafe wording before a person sees it.
Why we need Claude API credits
This work can't run without an LLM, and it never stops running
Answering a worried parent in their own language, turning a messy classroom note into a structured observation, and checking every draft for unsafe wording are language tasks. No rules engine or template does them well. They also run continuously: every question, every family, every week.
Credits go to the free community tier first. The educator rows show what the paid stage will add as it grows.
| Task | Model | Tokens per run | Runs / month | Tokens / month |
|---|---|---|---|---|
| Open Milestones · free community tier | ||||
| Answer a parent's question | Sonnet 5.5 | ~3k in + ~0.5k out = 3.5k | 3,000 × 8 = 24,000 | ≈ 84M |
| Safety check of each answer | Haiku 5.5 | ~1.5k in + ~0.1k out = 1.6k | 24,000 | ≈ 38.4M |
| Weekly play plan | Sonnet 5.5 | ~2.5k in + ~0.8k out = 3.3k | 3,000 × 4 = 12,000 | ≈ 39.6M |
| Translate one checklist age band into a new language | Sonnet 5.5 | ~20k in + ~15k out = 35k | 20 | ≈ 0.7M |
| Subtotal, free community tier | ≈ 162.7M | |||
| Classroom Observation Agent · educator pilot cohort | ||||
| Structure one child's daily notes | Haiku 5.5 + Sonnet 5.5 | ~4k in + ~1k out = 5k | 20 × 15 × 20 = 6,000 | ≈ 30M |
| Weekly parent update + safety audit | Sonnet 5.5 + Haiku 5.5 | ~6k in + ~1k out = 7k | 20 × 15 × 4 = 1,200 | ≈ 8.4M |
| Subtotal, educator pilot cohort | ≈ 38.4M | |||
| Total at launch-phase target | ≈ 201.1M tokens / month | |||
Projection at the 12-month target (estimate): 100,000 families reached, ~15,000 active in a given month × ~54k tokens per active family (8 × 3.5k + 8 × 1.6k + 4 × 3.3k) ≈ 810M, plus 100 classrooms × ~1.92M per classroom (300 × 5k + 60 × 7k) ≈ 192M. About 1.0B tokens per month, before translation work.
Built on Anthropic Claude
Why we build on Claude
Child development is a domain where nuance, memory and honesty matter more than speed. Claude Sonnet 5.5 is our reasoning and writing engine; Claude Haiku 5.5 handles high-volume triage and safety audits.
Reasoning for nuanced patterns
Development is rarely linear. Claude weighs context, like a new sibling, a bilingual home or a recent illness, before drawing anything from a single note.
Plain language, in many languages
Families should get the same quality of explanation whatever language they speak at home.
Calibrated, honest answers
We add citation checks and conservative escalation rules on top, so uncertain findings are flagged as uncertain, never invented.
observation.json
Sample output
{
"child_ref": "c_7f3a",
"age_months": 30,
"source": "teacher_note",
"milestone_refs": [
{ "domain": "language",
"item": "Combines words, incl. an action word",
"reference": "CDC · 30 mo",
"evidence": "said 'I want more crackers'" }
],
"uncertainty": "Single note; confirm over 2+ days",
"screening_flag": null,
"parent_update": { "lang": "en", "status": "awaiting_teacher_review" },
"model": "claude-sonnet-5-5"
}Children's data
What we promise families
Built for adults: parents, guardians and educators. Children never use Enfantz directly.
Data minimization
Open Milestones needs a child's age, not a name or photo. Names become pseudonymous IDs before any AI call.
Parents own it
Export or permanently delete a child's records at any time. Centers only see children whose parents consented.
No training on children's data
Never used to train ours or anyone's models. Anthropic's commercial API terms exclude inputs and outputs from training by default.
COPPA & GDPR-K aware
Designed to the US COPPA rule and GDPR rules for children's data from day one. Commitments, not certifications.
Roadmap
The next 24 months
Goals and targets, not results. Community first, then classrooms.
-
Now · months 0–3
Writing the Open Milestones checklist from public CDC and WHO guidance, building the Claude assistant and its safety reviewer, and preparing the open-source repository.
-
Next · months 3–12
Public launch, free and open source under MIT. Target by month 12: 100,000 families reached in 20 languages, plus a 20-classroom educator pilot cohort.
-
Later · months 12–24
The Classroom Observation Agent opens to every childcare center that wants it. Our goal: every parent, in every major language, gets the same quality guidance, and our reviewed milestone content becomes a public resource other nonprofits and clinics can reuse.
FAQ
Questions, answered honestly
Is this a diagnosis or medical advice?
No. Enfantz explains public milestone guidance in plain language and suggests play ideas. When a milestone is missed, it encourages you to talk with your pediatrician. It never ranks, scores or labels children.
Is Open Milestones really free?
Yes, free forever for individuals and families, with no ads and no selling of data. The paid product will be the Classroom Observation Agent for childcare centers.
Is my child's data used to train AI?
No. Information about children is never used to train our models or anyone else's. We use Claude through Anthropic's commercial API, whose terms exclude inputs and outputs from model training by default.
Does my child use the app?
No. Enfantz is for adults (18+): parents, guardians and educators. Children never interact with the AI.
When can I use it?
We're building it now. Join early access and we'll email you when the first version is ready to try.
Parents: get Open Milestones first.
Free forever, open source at launch, in your language.
Get early access