Languages

How do you build French AI apps with Plugsky?

French is close to English in structure but not in token cost: accented and elided forms such as l'utilisateur split into subwords, and French phrasing often runs longer. Plugsky serves it on the standard OpenAI-compatible endpoint — set base_url to https://api.plugsky.com/v1 and send UTF-8 text to /v1/chat/completions. Pin vous or tu, choose one regional variety, and use plugsky-embed-multilingual for mixed retrieval.

Key facts

French textUTF-8 Latin script with accents, ligatures such as œ and elisions such as l'utilisateur
Registervous and tu change verb forms throughout — keep one per surface
Regional varietiesFrance, Canada and African French differ in vocabulary and phrasing; evaluate the variety you serve
TokenisationAccented and elided forms can split into subwords — measure with the token calculator
API compatibilityOpenAI-compatible POST https://api.plugsky.com/v1/chat/completions — chat, streaming, JSON mode and function calling
Models30+ models behind one endpoint, from free tiers to frontier reasoning
Free tierFree plan with plugsky-micro and plugsky-lite, no card required
Product statusChat, streaming, JSON mode, function calling, embeddings, RAG and agents live; audio, images, moderation, files, batch, fine-tuning, assistants and responses coming soon

TL;DR

  • French runs on the standard endpoint with one base_url change.
  • Expect slightly more tokens than the English equivalent.
  • Choose vous or tu per surface and state it in the system prompt.
  • Keep accents in display and embedding text; normalise only for search keys.
  • plugsky-embed-multilingual covers French and English retrieval.

How it works, step by step

  1. Create a free Plugsky key and set base_url to https://api.plugsky.com/v1.
  2. Send real French prompts to /v1/chat/completions and compare two or three models.
  3. Measure tokens with the token calculator and size chunks from your own text, not translations of English samples.
  4. Decide vous or tu per surface and write it into the system prompt.
  5. For RAG, embed with plugsky-embed-multilingual and test French and English queries against one collection.
  6. Score outputs on a French gold set with native review before cutover.
1Create a freePlugsky key and setbase_url to2Send real Frenchprompts to/v1/chat/completions3Measure tokens withthe tokencalculator and size4Decide vous or tuper surface andwrite it into the5For RAG, embed withplugsky-embed-multilingualand test French and6Score outputs on aFrench gold setwith native review

Original data

UTF-8 Latin scFrench textOpenAI-compatiAPI compatibility30+ models behModelsSource: Plugsky facts table · updated 2026-09-26

Try it yourself

Open the token calculator →

How Plugsky handles French text

French needs no special endpoint: send UTF-8 text with accents, ligatures and apostrophes and the API returns the same. The engineering decisions sit in register and regional variety. vous and tu change verb forms across a whole sentence, so mixed prompts produce mixed outputs; and French written in France, Canada or West Africa differs in vocabulary, idiom and punctuation spacing.

Typography deserves a test too: French uses narrow no-break spaces before certain punctuation, and models sometimes omit them. If your product renders copy directly, add a normalisation step or accept the model's spacing.

Tokenisation and cost in French

French usually costs a little more per sentence than English because phrasing runs longer and accented or elided forms split into subwords. Apostrophes are the main fragmentation point: l'utilisateur, d'un, qu'il. Ligatures such as œ and accented capitals are rarer and usually handled, but worth checking in official text.

  • Measure tokens on native French, not on translated English samples.
  • Normalise apostrophe characters to one code point early.
  • Keep accents for display and embedding; fold them only in search keys.
  • Re-measure after changing models.

French retrieval and RAG

One multilingual collection built with plugsky-embed-multilingual serves French documents with French or English queries. Retrieval problems usually come from accents, apostrophes and regional vocabulary rather than from the embedding model itself.

  • Normalise apostrophes and whitespace before indexing.
  • Add accent-insensitive keys for keyword search alongside embeddings.
  • Keep regional variety consistent between documents and queries.
  • Test English queries against French documents explicitly.

Code example: a French request

Point your OpenAI client at https://api.plugsky.com/v1; only the content and system prompt change.

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": "Réponds en français avec le vouvoiement"}, {"role": "user", "content": "Résume ce contrat en trois points"}])

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

Honest comparison

CapabilityPlugskyFrench apps todayBuilding in-house
API compatibilityOpenAI-compatible — change base_url and model nameVaries by provider and SDKFull rewrite
French text handlingUTF-8 accents and elisions with vous/tu prompt controlDepends on provider tokeniser and prompt hygieneYou build normalisation and evals
Token budgetFixed tokeniser per model; measure native French and chunk to fitVaries by provider and modelYou host and tune each tokeniser
Multilingual retrievalplugsky-embed-multilingual for French and English RAGOften needs a separate embedding vendorYou serve and maintain embeddings
SovereigntyCloud, VPC, on-prem and air-gapped with residency optionsUsually US/EU public endpointsYou own the full stack

Frequently asked questions

Can Plugsky handle French text?

Yes. The API accepts UTF-8 French input, including accents, ligatures and elisions, on the OpenAI-compatible chat endpoint. Compare two or three models on your own prompts.

Why does French need more tokens than English?

French phrasing often runs longer and elided forms such as l'utilisateur split into subwords. Measure native French text with the token calculator before sizing context windows.

Should prompts use vous or tu?

vous for formal, B2B and service flows; tu for consumer and social products. State the choice in the system prompt and keep templates consistent.

Do I need to handle Canadian French separately?

Vocabulary and some phrasing differ. If you serve Canada, keep a separate evaluation slice and avoid averaging results with France French.

Is there a multilingual embedding model?

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

Can I keep data in the EU?

Plugsky supports cloud, VPC, on-prem and air-gapped deployment with residency options; confirm specifics with the docs and the enterprise team.

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.