FAQ

Who is Epithre for?

Teams building Indonesian-language products. We're the best fit if at least one of these matters to you:

Your users speak Bahasa Indonesia and quality matters.
Your data must stay in Indonesia (UU 27/2022 compliance).
You want predictable, low pricing for high-volume workloads.
You want OpenAI-wire-compatible API so your existing SDK code works without modification.

What makes Epithre distinct?

Indonesian-tuned models: native register handling, code-switching, domain idiom (legal / medical / finance) without prompt-gymnastics.
Jakarta-hosted: low latency for Indonesia traffic, hard data-residency guarantee.
Predictable pricing: per-token rates published on the pricing page; no negotiated tiers or hidden fees.
Cross-modal embeddings: text and image in one 4000-dim vector space, directly cosine-comparable for cross-modal retrieval.
Built-in rerank: cross-encoder rerank in the same platform, no second vendor needed.

Is Epithre production-ready?

Currently in alpha. Wire format is stable, features are real and shipping, pricing is settled. What's still maturing: SLA commitments, multi-region failover, full enterprise compliance audits.

Suitable for: dev / staging environments, MVPs, production for internal tools, production for apps where some downtime is acceptable.

Not yet recommended for: anything where downtime causes legal / financial / safety consequences. Email us; we can discuss enterprise-tier arrangements with SLA.

What's the SLA?

During alpha: best-effort. Realistic uptime: 99.5%-99.8%. We don't yet offer credits for downtime.

For enterprise: we can discuss formal SLA (99.9% with credits) on a per-contract basis.

Do you offer fine-tuning?

Not self-serve yet. We do offer custom training as a managed service via email:

You send us your training data and desired model.
We rent H100 cloud compute and train a LoRA adapter.
We deploy the adapter to our inference stack, served via a custom model ID.
You pay Rp3,500,000-9,000,000 per training run + per-token inference at the base model's rate.

Email hello@epithre.com with subject "Custom model" to discuss.

Can I host my data on-premises?

Not currently. Epithre is a managed service. We can discuss an enterprise on-prem deployment for very large customers - email us.

What models do I get?

Model	What it is
`epithre-omni`	Flagship multimodal chat (vision, tools, thinking)
`epithre-prme`	180K-context premium for long docs and codebases
`epithre-lyt`	Fast cheap chat (image + audio + video input)
`epithre-embed`	4000-dim multimodal embeddings (text + image, shared space)
`epithre-rerank`	Cross-encoder rerank, standard `/rerank` body shape
`epithre-iris`	Image generation + editing

Audio (Whisper-equivalent) and TTS not yet offered.

How do I migrate from OpenAI?

Three lines of config change: api_key, base_url, model ID. Wire format is identical. See migration guide.

What languages does it speak?

Strongest: Bahasa Indonesia (native), English (very strong).

Reasonably good: most major SEA languages (Tagalog, Vietnamese, Thai, Malay), Chinese, Arabic.

Weaker: low-resource languages (regional Indonesian dialects, indigenous languages).

Can I use it for personal projects?

Yes. Rp50,000 free credit on signup; no minimum spend; no commitment. Just sign up at platform.epithre.com.

What's your refund policy?

Unused credit refunded on account closure, minus payment processor fees. Email hello@epithre.com for refund requests.

How do I get help?

Documentation: this site (docs.epithre.com).
Email: hello@epithre.com (response within 24 hours during alpha; 4-8 hours during business hours).
Status page: planned. Will be at status.epithre.com.
Community: we don't have a Discord / forum yet. Direct email is the channel.

FAQ