There are jokes that only make sense in Javanese. Teasing words that feel best in Sundanese. A lullaby in Bugis that somehow says “I love you” better than any Indonesian sentence ever could.

Problem is, those moments (full of warmth, richness, familiarity) are becoming rarer. In many homes, the local language has quietly moved from everyday speech to family memory. A phrase here. A proverb there.
We miss the tone, the cadence, the way our grandparents spoke — not just what they said, but how they said it.  

Now, imagine if AI could help. Not to replace those voices, but to carry them forward.

That’s the promise behind Sahabat-AI, Indonesia’s own large language model that is recently updated with a new 70-billion-parameter architecture. Built on LLaMA 3.1 and fine-tuned with Indonesian and regional languages like Javanese, Sundanese, Balinese, and Batak Toba, this version deepens its cultural fluency. It ranks among the top in Southeast Asia on benchmarks like SEA-HELM and IndoMMLU, and can now run efficiently on minimal infrastructure.

As we evaluate the progress of other prominent LLM companies, like OpenAI, Gemini, and Deepseek, it's not surprising to see the roadmap for Sahabat-AI, which includes the introduction of a reasoning model and an agentic platform in its upcoming release. Given the trajectory of industry trends, Sahabat-AI is poised to align with these advancements, offering more sophisticated capabilities and pushing the boundaries of AI outcomes driven by agents.

While much of the AI conversation globally has focused on efficiency like automating work or accelerating content, here, I’m are asking something different:

Can AI help us remember what makes us feel at home?

In New Zealand, a project called Te Hiku Media built a Māori language model trained on hours of community-donated speech, not just to document the language, but to keep it alive through tools that speak and listen like a native. The lesson is clear: AI doesn't have to mean cultural erasure. It can mean digital continuity if built with respect and inclusion.

I want to share my take on reimagining the 'what ifs' with Sahabat AI:

I imagine an AI that remembers our custom when we begin to forget.

A friend gets married in Jakarta. Her parents are from Padang. Her in-laws, from Flores. The question isn’t just what colour the dress should be — it’s whose version of the ceremony do we follow? What do we say? Who leads it? What matters most? What is even the value of all this?

Usually, someone calls an elder. Or asks in the family group chat. Sometimes they guess. Sometimes, they skip it entirely because no one understand its significance.  

Now imagine opening Sahabat-AI and whispering a quiet, respectful question:
"What are the steps in a Minangkabau marriage ceremony? What do the symbols mean?"

And the AI, which trained not just on information but intention as well, tells you the whole story, the purpose, the path, everything. It doesn’t prescribe. It suggests, with care. With authoritative citation. With humility and perceived empathy of your conditions.

This Ritual & Ceremony Vault could become an intergenerational bridge, especially for those living far from home, or born in mixed cultures, or simply too afraid to ask. It could help preserve rituals without freezing them, making space for both tradition and transition.

But here, there’s a quiet risk: not everything should be documented.

Some rituals are sacred, others are private, or evolving, and certain custom has a price that is too much to bear in this time. If we reduce them to scripts, we lose the spirit. So again, the challenge isn’t technical. It’s contextual. The AI must earn its place through consent, collaboration, and cultural guardrails.

Still, when we built with respect, it could become a living archive of us. A way to ensure that what binds us in life, and honours us in death, doesn’t get lost to time or distance.  

I imagine an AI that can laugh with you in Sundanese, and every Indonesia language we grow with.

AI can simply translate jokes. But most take the life of out of it. We need an AI that really get it.
The kind of AI that understands why a Sundanese joke lands best when it comes with a naive and nonchalant facade. Why "kumaha damang?" and a smile isn’t just a question but a social dance.  

Sundanese, like many regional languages, carries more than vocabulary. It carries manner and value. Softness in speech. Humor as politeness. Euphemisms as code. It’s how you flirt without offense, reject without confrontation, tease without tension. These aren’t things you can look up in a dictionary, they live in conversation, passed down through tone, timing, and trust. Tone, especially.

But as the everyday use of Sundanese shrinks — especially in urban areas and younger generations — we’re losing more than words. We’re losing a whole way of being.

That’s why I imagine an everyday language companion built for fluency in feeling.

An AI that you can chat with like you would an older cousin — who not only replies in proper Sundanese, but teaches you when you get it wrong, explains why some phrases feel warmer than others, nudges you to use the lemes register when talking to elders, and throws in a joke when the moment’s right.

It could sit quietly in your phone, helping you write a birthday wish for your grandmother in proper Sundanese. Or guide you in how to respond to a sarcastic "euleuh-euleuh" without sounding awkward. Or translate your urban slang into a version your village uncle would find charming.

Of course, it’s not without risk.

Get it slightly wrong — the tone, the context, the timing — and it can quickly feel artificial, even offensive. A flat joke in a lively language sting more than silence. Cultural trust isn’t earned by data scraping. It is earned by showing respect, accuracy, and care.

But if we do it right with communities, not just coders, it could become something truly precious:
A companion that helps you keep your language alive not as a subject to study, but as a way to live.
A small, ever-available reminder that your voice still belongs in your own language.

Because sometimes, what we miss most isn’t just the words. It is the laughter that only makes sense when it’s said in our local language that dominates the tongue when we grow up.

A Way to Remember, A Reason to Continue

If language is how we pass down memory, and ritual is how we hold onto meaning, then tools like Sahabat-AI shouldn’t just be built for productivity, but for presence. For the small, soft ways we stay connected to where we’re from, who we are, and how we once spoke to each other when no one was watching.

If that is too personal, let’s take it higher. In Indonesia, where over 700 languages are spoken, one of the biggest barriers to education and access is language diversity. By breaking down language barriers, Sahabat-AI opens up new opportunities for inclusive education, allowing students, educators, and communities across Indonesia to access quality learning resources in their native languages. Additionally, its ability to understand cultural nuances makes it a valuable tool in daily life, community services, and beyond, ensuring that the wealth of knowledge we have isn’t limited by language, but shared across the nation

This kind of AI that, if we let it, can live alongside us. Not to interrupt, but to accompany: in our chats, in our celebrations, in the questions we’re too embarrassed to ask out loud. It could sit quietly in our keyboards, or guide us through a wedding plan, or help us write a message we wish we could still say to someone who's gone.

But it can’t do this alone. Not without the people who carry the knowledge. Not without the communities who know what’s worth keeping. This is a project that will need curators, speakers, elders, artists, learners — those who understand that culture is not what we store, but what we continue to use.

Eventually, perhaps, it becomes more than a tool. Maybe it becomes a way to immortalize what matters, not in stone or script, but in something more alive:

A shared way of speaking.

A pattern of being.

A quiet presence that helps us hold onto the parts of ourselves we’re not ready to let go.

Enjoyed the read? Let’s take it further.

Connect to unlock exclusive insights, smart AI tools, and real connections that spark action.

Schedule a chat to unlock the full experience