canonical: https://jentic.com/apis/anyscale.com/anyscale # Anyscale Endpoints API Jentic publishes the only available OpenAPI specification for Anyscale Endpoints API, keeping it validated and agent-ready. The Anyscale Endpoints API serves open-source large language models through an OpenAI-compatible interface, so existing chat and completion clients work with minimal change. It generates chat completions and text completions, produces embeddings for retrieval and similarity tasks, and lists the models available for inference. This gives agents hosted access to open models without running the serving infrastructure themselves. ## For AI agents Generate chat and text completions from open-source LLMs, create embeddings, and list available models through an OpenAI-compatible interface. ## Scope Does not fine-tune, train, or host custom models, and does not manage Anyscale compute clusters. Use for inference against the served open models only. ## Capabilities - Generate a chat completion from an open-source language model - Generate a text completion for a prompt - Create embeddings for text used in retrieval or similarity search - List the models available for inference ## Use cases ### Open-Model Chat Completions An agent needs responses from an open-source LLM rather than a proprietary one, for cost or licensing reasons. Anyscale exposes chat completions through an OpenAI-compatible schema, so the agent submits messages and reads back the assistant reply. Through Jentic the agent searches for the chat operation and calls it directly with its bearer credential managed for it. Example prompt: Send a list of chat messages to an open-source model and return the assistant's reply ### Embeddings for Retrieval A retrieval pipeline converts documents and queries into vectors before searching a vector store. Anyscale's embeddings operation returns vectors for input text so the pipeline can index and match on semantic similarity. Using open embedding models keeps the vector space consistent and self-contained. Example prompt: Create embedding vectors for a batch of text passages for indexing in a vector store ### Model Discovery Before Inference Before selecting a model, an agent lists what Anyscale currently serves so it can pick one that matches the task and context window. The models operation returns the available identifiers, which the agent then passes to a completion call. This avoids hard-coding a model name that may change. Example prompt: List the models Anyscale serves and select one for a chat completion request ### Drop-In Replacement for OpenAI Clients A team already using an OpenAI-compatible client wants to route some traffic to open models. Because Anyscale mirrors the chat and completion schemas, the same client works by pointing at the Anyscale base URL. Through Jentic the endpoint is imported once and the credential is stored, so switching providers is a configuration choice. Example prompt: Route a chat completion request to an Anyscale open model using an OpenAI-compatible payload ## Key endpoints | Method | Path | Description | | --- | --- | --- | | POST | `/chat/completions` | Generate a chat completion from an open-source model | | POST | `/completions` | Generate a text completion for a prompt | | POST | `/embeddings` | Create embedding vectors for input text | | GET | `/models` | List the models available for inference | ## Key resources - **Chat completions** — Generate assistant replies from a list of chat messages - **Completions** — Generate a text completion for a prompt - **Embeddings** — Produce embedding vectors for input text - **Models** — List the open-source models available for inference ## Why Jentic - **Setup:** Wiring Anyscale by hand means managing its bearer token against api.endpoints.anyscale.com and pointing an OpenAI-compatible client at the right base URL yourself. Through Jentic you install once, import Anyscale from the API Directory, store the credential once, and your agent calls it. - **Permission scoping:** Anyscale's operations are inference calls, so a rule limits your agent to the operations it needs, such as chat completions and embeddings. You choose which operations it may call, so text completions or model listing are included only when you add them. - **Credential handling:** Your Anyscale 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 'generate a chat completion' or 'create embeddings', and Jentic returns the matching Anyscale operation with its input schema so the agent calls the right endpoint without browsing the reference docs. ## Related APIs - **OpenAI API** — OpenAI serves proprietary models through the same chat and embeddings shapes Anyscale mirrors for open models. - **Mistral AI** — Mistral offers its own open and commercial models through a hosted API, an alternative source of inference. - **Cohere** — Cohere provides text generation and embeddings with a retrieval focus, overlapping Anyscale's embeddings use. - **Hugging Face** — Hugging Face hosts the open model weights and cards; Anyscale serves many of those models for low-latency inference. ## FAQ ### Why is there no official OpenAPI spec for Anyscale Endpoints API? Anyscale does not publish an OpenAPI specification for its endpoints. Jentic generates and maintains this spec so that AI agents and developers can call the Anyscale Endpoints API via structured tooling. It is validated against the live API and kept up to date. To run it on your own infrastructure, install Jentic One from its GitHub repo. ### What authentication does the Anyscale Endpoints API use? The Anyscale Endpoints API authenticates with an HTTP bearer token in the Authorization header, matching the OpenAI-compatible convention. Through Jentic, that token is stored encrypted by your own Jentic One instance and injected at execution time, so it never enters the agent's prompt or logs. ### Is the Anyscale Endpoints API compatible with OpenAI clients? Yes. Anyscale mirrors the OpenAI chat completions, completions, and embeddings schemas, so an OpenAI-compatible client works by pointing at the Anyscale base URL. This lets an agent route requests to open-source models without rewriting its request payloads. ### What are the rate limits for the Anyscale Endpoints API? The OpenAPI spec does not specify rate limits for the Anyscale Endpoints API. Check the Anyscale documentation at https://docs.anyscale.com for current limits, and back off when you receive a 429 response. ### How do I generate a chat completion through Jentic? Search Jentic for 'generate a chat completion'. Jentic returns Anyscale's chat operation with its input schema covering the model name and message list, so your agent can call it directly and read back the assistant reply. ### Can I limit what my agent is allowed to do with the Anyscale Endpoints API? Yes. You write a rule that allows only the operations the agent needs, so generating embeddings can stay separate from chat completions unless you allow both. Because Jentic One is self-hosted, those rules and the audit log of every call stay on your own infrastructure. ### Is there an Anyscale Endpoints API MCP server? You do not need an MCP server to give your agent the Anyscale Endpoints API. Jentic connects it directly from the API Directory: import it, store your credential once, and your agent can generate completions and embeddings. That also keeps your agent's context free of an extra server's tool definitions.