How do you build Norwegian AI apps with Plugsky?

Key facts

Written standards	Bokmål and Nynorsk are both official; most digital content is Bokmål but public bodies may require both
Norwegian script	Latin script with æ, ø and å; UTF-8 handled as-is
Compounds	Compound nouns split into several subwords — measure with the tokenizer
Register	du is standard in modern usage; De survives only in very formal or ceremonial text
API compatibility	OpenAI-compatible POST https://api.plugsky.com/v1/chat/completions — chat, streaming, JSON mode and function calling
Models	30+ models behind one endpoint, from free tiers to frontier reasoning
Free tier	Free plan with plugsky-micro and plugsky-lite, no card required
Product status	Chat, streaming, JSON mode, function calling, embeddings, RAG and agents live; audio, images, moderation, files, batch, fine-tuning, assistants and responses coming soon

TL;DR

• Norwegian works on the standard endpoint with one base_url change.
• Choose Bokmål or Nynorsk per product; do not mix them silently.
• Compounds are the main token variable — count domain vocabulary.
• Use du for modern products; De reads ceremonial.
• plugsky-embed-multilingual covers Norwegian and English retrieval.

How it works, step by step

1. Decide Bokmål or Nynorsk per surface, and whether public-sector rules require both.
2. Create a free Plugsky key and set base_url to https://api.plugsky.com/v1.
3. Send compound-heavy Norwegian text to /v1/chat/completions and compare two or three models.
4. Measure tokens with the tokenizer and leave headroom for compound-dense documents.
5. For RAG, embed with plugsky-embed-multilingual and test Norwegian queries against English documents.
6. Review outputs with native speakers on a fixed gold set, then move production traffic.

Original data

Try it yourself

Open the tokenizer →

How Plugsky handles Norwegian text

The API treats Norwegian like any UTF-8 language: no special endpoint, no locale flag. The decision that matters is written standard. Bokmål dominates digital text and business use; Nynorsk is official alongside it and required in parts of the public sector. Mixing them within one product reads as an error to Norwegian readers, so choose deliberately.

Register is simpler than in many languages: du is the normal address form in modern Norwegian, including business communication, and De appears only in deliberately formal or ceremonial contexts. Keep that policy in the system prompt rather than leaving it to the model.

Tokenisation and cost in Norwegian

Everyday Norwegian is close to English in token cost, but compounds are written joined and split into subwords in the tokeniser. Public-sector, legal and technical text compounds heavily, and number words are long and frequent in invoices and reports.

• Measure on the text you actually process, not on chat samples.
• Keep compounds intact for display; split for search keys only.
• Test abbreviations common in Norwegian official writing.
• Re-count after switching models or tokenisers.

Norwegian retrieval and RAG

One multilingual collection handles Norwegian documents with Norwegian or English queries when you embed with plugsky-embed-multilingual. Retrieval issues usually come from compound variants and from Bokmål/Nynorsk mismatch — a query in one standard against documents in the other.

• Add compound-splitting or synonym keys for keyword search.
• Test Bokmål and Nynorsk queries against both document sets.
• Include English queries, which are common in technical teams.
• Keep one embedding model per collection.

Code example: a Norwegian request

Point your OpenAI client at https://api.plugsky.com/v1; the request shape is unchanged from any other language.

client = OpenAI(base_url="https://api.plugsky.com/v1", api_key=os.environ["PLUGSKY_API_KEY"])

client.chat.completions.create(model="plugsky-pro", messages=[{"role": "system", "content": "Svar på norsk bokmål med en hyggelig, profesjonell tone"}, {"role": "user", "content": "Skriv et svar til kunden om den forsinkede leveransen"}])

Start on the free plan with plugsky-micro and plugsky-lite, then compare paid models on a Norwegian gold set. See the docs for the API reference.

Honest comparison

Capability	Plugsky	Norwegian apps today	Building in-house
API compatibility	OpenAI-compatible — change base_url and model name	Varies by provider and SDK	Full rewrite
Written standards	Bokmål or Nynorsk via prompt policy and separate evaluation	Depends on provider training mix	You curate data per standard
Token budget	Fixed tokeniser per model; measure compounds and chunk to fit	Varies by provider and model	You host and tune each tokeniser
Multilingual retrieval	plugsky-embed-multilingual for Norwegian and English RAG	Often needs a separate embedding vendor	You serve and maintain embeddings
Sovereignty	Cloud, VPC, on-prem and air-gapped with residency options	Usually US/EU public endpoints	You own the full stack

Frequently asked questions

Can Plugsky handle Norwegian text?

Yes. The API accepts UTF-8 Norwegian input, including æ, ø and å, on the OpenAI-compatible chat endpoint. Compare models on your own prompts before choosing.

Bokmål or Nynorsk?

Pick per product and audience. Most digital and business content is Bokmål, while parts of the public sector require Nynorsk. State the standard in the system prompt and evaluate separately.

How do compounds affect token usage?

Joined compound nouns split into several subwords, so a single word can cost multiple tokens. Measure on domain text with the tokenizer and leave context headroom.

Is du or De correct?

du is the standard form in modern Norwegian, including business use; De is reserved for very formal contexts. Put the choice in the system prompt.

Is there a multilingual embedding model?

Yes — plugsky-embed-multilingual is part of the 30+ model catalogue and supports Norwegian and English retrieval from one collection.

Can I deploy in my own environment?

Yes. Plugsky supports cloud, VPC, on-prem and air-gapped deployment with residency options for regulated teams.

Is there a free plan?

Yes — plugsky-micro and plugsky-lite are free with no card, and a 14-day full-access trial covers the paid catalogue.