Build with EPITHRE. API

AI inference, tuned for Indonesia. Chat, embeddings, reranking, and image generation with native Bahasa Indonesia fluency. Data residency in Jakarta. OpenAI-wire-compatible, so you can drop it into your existing SDK in 30 seconds.

What you get

• OpenAI-compat wire format. Use the official openai SDK (Python, Node, Go, Java, Ruby, PHP, Rust). Just change base_url to https://api.epithre.com/v1.
• Six model tiers under one API: epithre-omni (flagship multimodal chat), epithre-prme (premium long context), epithre-lyt (compact), epithre-embed (multimodal embeddings), epithre-rerank (cross-encoder rerank), epithre-iris (image generation + editing).
• Native Indonesian quality. Tuned on Indonesian corpora; code-switching, formal vs casual register, and Indonesian legal/finance/medical idioms work without prompt-gymnastics.
• Data residency Jakarta. All inference and storage on hardware in Indonesia. No data egress under normal operation.
• Ecosystem features: prompt caching, structured output, batch API (50% off), webhooks, knowledge-document upload + retrieval, cross-modal embed in one shared vector space.

60-second quickstart

from openai import OpenAI

client = OpenAI(
    api_key="esk_live_...",                  # from platform.epithre.com -> API Keys
    base_url="https://api.epithre.com/v1",
)

resp = client.chat.completions.create(
    model="epithre-omni",
    messages=[{"role": "user", "content": "Halo, apa kabar?"}],
)
print(resp.choices[0].message.content)

Full quickstart covers the four 1-minute walkthroughs (chat, embeddings, image, batch). Authentication covers API keys and JWT sessions for the dashboard.

What to read next

If you're coming from another provider, jump to the migration guide:

• Migrate from OpenAI
• Migrate from Anthropic Claude
• Migrate from Cohere

If you're starting fresh, the reference pages are the truth source for every parameter:

• Chat completions, Embeddings, Rerank
• Image generation + edit
• Files, Batches, Retrieval, Webhooks

If you want to write good prompts for Indonesian content (the part most other providers get wrong), read the prompt engineering guide.

Why Epithre exists

For Indonesian-language workloads, three things compound to make a purpose-built local platform the right choice:

1. Quality: Epithre models are tuned on Indonesian corpora. Idiom, code-switching, formal vs casual register, and domain language (legal, medical, finance) work without prompt-gymnastics.
2. Latency: a request from Jakarta inside Jakarta round-trips under 30ms before inference. Cross-continent round-trips add 250-300ms before any work begins.
3. Compliance: under UU 27/2022 (Pelindungan Data Pribadi), some workloads require data not to leave Indonesia. Our residency is a hard guarantee, not an option.

We focus on being the best-priced, lowest-friction, fastest-for-Jakarta API for teams shipping Indonesian-language products.

Status

Currently in alpha. The wire format is stable. The features are real and shipping. Pricing is settled. What's still in motion: SLA commitments, multi-region failover, full enterprise compliance audits. Email us if you need any of those today; we'll work something out.

• Live: API gateway, dashboard, docs, all endpoints listed in the reference.
• Best-effort SLA during alpha. Realistic uptime: 99.5%-99.8%.
• Manual top-up via email during alpha. Stripe/Midtrans integration is planned.

Updates ship to the changelog. Subscribe to release notifications via webhook platform.update event (coming soon) or follow hello@epithre.com.