canonical: https://jentic.com/apis/aelph-alpha.com/aleph-alpha # Aelph Alpha Aleph Alpha PhariaAI API Jentic publishes the only available OpenAPI specification for Aleph Alpha PhariaAI API, keeping it validated and agent-ready. PhariaAI is the inference platform behind Aleph Alpha's Luminous family of large language models, exposing endpoints for text completion, chat, embeddings, semantic similarity, evaluation, question answering, summarisation, and tokenisation. The API is European-hosted and aimed at enterprise teams that need sovereign LLM infrastructure for retrieval, classification, and generation workloads. Authentication uses a bearer token issued from the Aleph Alpha account portal. ## For AI agents Run text completion, chat, embeddings, summarisation, Q&A, and tokenisation against Aleph Alpha's Luminous LLMs from a single European-hosted inference API. ## Scope Does not handle model fine-tuning, training data uploads, or non-text modalities - use for inference against pre-trained Luminous models only. ## Capabilities - Generate text completions and chat responses with Luminous-family models - Produce dense and semantic embeddings for retrieval and similarity workflows - Run reference-based answer evaluation for grading and guardrails - Answer questions over supplied context documents - Summarise long documents into structured outputs - Tokenise and detokenise text to manage prompt budgets and audit token usage ## Use cases ### Sovereign Enterprise RAG Pipeline Build retrieval-augmented generation on infrastructure that stays inside the EU. PhariaAI's /semantic_embed endpoint produces embeddings for indexing, /qa answers questions with supplied context, and /complete or /chat/completions handle final generation. Suitable for regulated industries that need an LLM stack outside US-controlled clouds while still operating in English, German, French, Spanish, and Italian. Example prompt: Call POST /semantic_embed for 500 policy documents, store the vectors in pgvector, then call POST /qa with the top-k passages to answer a user question ### Document Summarisation at Scale Summarise contracts, research papers, or customer transcripts using the dedicated /summarize endpoint. The endpoint accepts a document and returns a structured summary, removing the need to hand-craft summarisation prompts for the /complete endpoint. Pair with /tokenize to split documents that exceed the model context window before summarising chunk-by-chunk. Example prompt: Call POST /tokenize to chunk a 50-page report, then call POST /summarize on each chunk and concatenate the bullet outputs ### Classification and Grading with Evaluate Use /evaluate to score how well a candidate completion matches a reference, which is the basis for classification, grading, and answer-verification workflows. Combined with /qa for retrieval and /complete for free-form generation, this covers the core enterprise NLP loop on a single European inference platform. Example prompt: Call POST /evaluate with a candidate answer and a list of reference answers and return the log-probability score for each reference ### AI Agent with Sovereign LLM Backend Route an AI agent's reasoning through Luminous instead of a US-hosted model when data residency matters. Through Jentic, the agent searches by intent for the right PhariaAI operation, loads its schema, and executes with the bearer token isolated in the vault. The Jentic value chain stays the same - search, load, execute - but the model call lands on European infrastructure. Example prompt: Search Jentic for 'generate a chat completion with luminous', load the /chat/completions schema, and execute with the user's messages and model=luminous-supreme-control ## Key endpoints | Method | Path | Description | | --- | --- | --- | | POST | /complete | Generate a text completion | | POST | /chat/completions | Generate a chat completion from a message history | | POST | /embed | Create dense embeddings for one or more texts | | POST | /semantic_embed | Create semantic embeddings tuned for similarity search | | POST | /qa | Answer a question using supplied context | | POST | /summarize | Summarise a document | | POST | /evaluate | Score a completion against reference answers | | GET | /models_available | List Luminous models available to the account | ## Key resources - **Completion** — Text generation via /complete and /chat/completions - **Embeddings** — Dense and semantic embeddings via /embed and /semantic_embed - **Q&A** — Context-grounded question answering via /qa - **Summarization** — Document summarisation via /summarize - **Evaluation** — Reference-based scoring via /evaluate - **Tokenization** — Tokenise and detokenise text via /tokenize and /detokenize - **Models** — List available Luminous models via /models_available - **Account** — Check remaining credit via /users/me/tokens ## Why Jentic - **Setup:** Wiring the Aleph Alpha PhariaAI API by hand means learning its bearer auth and coding calls against several inference endpoints. Through Jentic you install once, import the PhariaAI API from the API Directory, store the token once, and your agent calls it. - **Permission scoping:** PhariaAI takes the model and prompt in the request body across inference operations, so limit the agent to the operations it needs, such as summarize or embed. You choose which operations are in the allowed set, so completion or evaluation is not reachable unless you add it. - **Credential handling:** Your Aleph Alpha bearer token is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context. - **Discovery method:** Agents search Jentic by intent such as 'summarise a long document with Luminous', and Jentic returns the matching PhariaAI operation with its input schema so the agent picks summarize over complete without browsing the reference docs. ## Related APIs - **OpenAI API** — Largest commercial LLM provider with GPT models hosted in the US - **Cohere** — Enterprise LLM provider with strong embedding and rerank endpoints - **Mistral AI** — European LLM provider with open-weight and hosted Mistral models - **Hugging Face** — Model hub and inference API across thousands of models ## FAQ ### Why is there no official OpenAPI spec for Aleph Alpha PhariaAI API? Aleph Alpha does not publish an OpenAPI specification. Jentic generates and maintains this spec so that AI agents and developers can call Aleph Alpha PhariaAI API via structured tooling. It is validated against the live API and kept up to date. Get started with Jentic One, the self-hosted execution layer. ### What authentication does the Aleph Alpha PhariaAI API use? PhariaAI uses HTTP bearer authentication; pass an Authorization: Bearer header with a token created in the Aleph Alpha account portal. Through Jentic the token is stored in the vault and applied at execution time. ### Which models are available through the PhariaAI API? Call GET /models_available to list the Luminous variants - base, extended, supreme, supreme-control, and any newer models - provisioned for your account. The same model identifier is then passed in the body of /complete, /chat/completions, /embed, and other endpoints. ### Can I generate embeddings for semantic search with PhariaAI? Yes. POST /embed returns dense embeddings suitable for general use, and POST /semantic_embed returns embeddings tuned for semantic similarity. Both accept arrays of texts and return float vectors that can be stored in any vector database. ### What are the rate limits for the PhariaAI API? Rate limits are tied to account tier and are not declared in the spec. Check your Aleph Alpha account dashboard or call GET /users/me/tokens to see remaining credit; HTTP 429 responses are surfaced through Jentic so agents can back off. ### How do I run a chat completion through Jentic? Run pip install jentic, search Jentic with 'generate a chat completion with luminous', load the /chat/completions schema, and execute with model and messages in the request body. Jentic injects the bearer token at execution time. ### Can I limit what my agent is allowed to do with the Aleph Alpha PhariaAI API? Yes. Because Jentic One is self-hosted, you decide which PhariaAI operations sit in the allowed set, so you can restrict an agent to just /summarize and /embed while leaving /complete, /chat/completions, and /evaluate unreachable. Every inference call takes the model and prompt in the request body, so scoping happens at the operation level: an operation the agent cannot reach cannot be invoked no matter what prompt it receives. Your bearer token is applied only for the operations you permit, keeping the agent inside the boundary you set.