canonical: https://jentic.com/apis/mixedbread.com/mixedbread # Mixedbread API The Mixedbread API is a retrieval and search platform for AI applications. It generates text embeddings, reranks candidate results against a query, and runs vector stores that index your documents for semantic search and question answering. It also parses documents into clean text and extracts structured fields from them, so an agent can build the full pipeline from raw files to retrievable knowledge. Mixedbread targets teams building retrieval-augmented generation and semantic search over their own content. ## For AI agents Generate text embeddings, rerank results against a query, run vector stores for semantic search and question answering, and parse and extract structured data from documents. Authenticates with a Bearer token in the Authorization header. ## Scope Does not handle model fine-tuning, image generation, or billing administration. Use for embeddings, reranking, vector stores, and document parsing and extraction only. ## Capabilities - Generate embedding vectors for text inputs - Rerank a list of candidate documents against a query - Create and manage vector stores that index documents for retrieval - Run semantic search and question answering over a vector store - Parse documents into clean text and extract structured fields from them ## Use cases ### Build a retrieval pipeline for RAG An AI agent can embed a corpus and index it in a Mixedbread vector store, then retrieve the most relevant passages at query time to ground a language model's answer. It generates embeddings for the text, creates a store, and runs a search when a question arrives. Through Jentic the agent discovers each operation by intent and calls it with the text and query. Example prompt: Generate embeddings for a set of passages, create a vector store, and return the store identifier ### Rerank candidate results for better relevance When a first-pass search returns many candidates, an agent reranks them against the original query so the most relevant results rise to the top before they reach the model or the user. This improves answer quality without re-indexing the corpus. The reranking call takes the query and the candidate list and returns them ordered by relevance. Example prompt: Rerank a list of ten candidate passages against a user query and return them ordered by relevance ### Turn documents into retrievable knowledge An agent can parse uploaded documents into clean text and extract structured fields from them, then embed and index the result for search. This lets a team point Mixedbread at raw files and get back a queryable knowledge base. Parsing and extraction run as jobs the agent can submit and read back. Example prompt: Submit a parsing job for an uploaded document and return the job id so the parsed text can be retrieved ## Key endpoints | Method | Path | Description | | --- | --- | --- | | POST | `/v1/embeddings` | Generate embedding vectors for text inputs | | POST | `/v1/reranking` | Rerank candidate documents against a query | | POST | `/v1/stores` | Create a vector store | | POST | `/v1/stores/search` | Search a vector store for relevant passages | | POST | `/v1/parsing/jobs` | Submit a document parsing job | | POST | `/v1/extractions/jobs` | Submit a structured extraction job | ## Key resources - **Embeddings** — Generate embedding vectors for text inputs - **Reranking** — Reorder candidate documents by relevance to a query - **Stores** — Create and manage vector stores and run search and question answering over them - **Parsing and extraction** — Parse documents into text and extract structured fields as jobs ## Why Jentic - **Setup:** Wiring the Mixedbread API by hand means setting up its Bearer token auth, formatting the embedding, reranking, store, and job calls, and polling parsing and extraction jobs to completion yourself. Through Jentic you install once, import the Mixedbread API from the API Directory, store the key once, and your agent calls it. - **Permission scoping:** The Mixedbread API addresses each vector store by an identifier in the URL path, so you can scope the agent to the stores and operations it needs: allow it to generate embeddings and search a store and leave store deletion out of the allowed set unless you add it. - **Credential handling:** Your Mixedbread API key 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 embeddings' or 'search my documents', and Jentic returns the matching Mixedbread operation with its input schema so the agent calls the right endpoint without reading the reference docs. ## Related APIs - **Cohere API** — Also offers embedding and rerank models versus Mixedbread's combined embeddings, rerank, and vector stores. - **OpenAI API** — Provides embeddings alongside its language models but leaves reranking and vector storage to you. - **Pinecone API** — A dedicated vector database that can store the embeddings Mixedbread generates. ## FAQ ### What authentication does the Mixedbread API use? Mixedbread authenticates with a Bearer token, your Mixedbread API key, sent in the Authorization header of every request. Through Jentic the key is stored encrypted by your own Jentic One instance and supplied to requests at execution time, so the agent never sees the raw value. ### Can I generate embeddings and rerank results with the Mixedbread API? Yes. One operation returns embedding vectors for text inputs, and another reranks a list of candidate documents against a query so the most relevant rise to the top. An agent connected through Jentic can call both to build and refine a retrieval pipeline. ### How do vector stores work in the Mixedbread API? You create a vector store, add documents to it, and then run search or question answering over the store. The store indexes your content so an agent can retrieve the most relevant passages at query time without managing its own vector database. ### What are the rate limits for the Mixedbread API? The OpenAPI spec does not declare rate limits for the Mixedbread API. Apply conservative client-side throttling, back off on HTTP 429 responses, and check the Mixedbread developer documentation for the current published limits. ### Can I limit what my agent is allowed to do with the Mixedbread API? Yes. You choose which operations the agent may call, so it can be allowed to generate embeddings and search a store while store deletion stays out of the allowed set unless you add it. Because each store is addressed by an identifier in the URL path, you can also keep the agent to the stores it needs, and every call it makes is logged. ### Is there a Mixedbread API MCP server? You don't need an MCP server to give your agent Mixedbread. Jentic connects it directly from the API Directory: import it, store the key once, and your agent calls the embedding, reranking, and search operations on demand without a separate server to run. ### How do I build a retrieval pipeline with Mixedbread through Jentic? Search Jentic for an intent like 'generate embeddings' or 'search my documents', which returns the matching Mixedbread operation with its input schema so the agent supplies the text or query and reads the response. To run it on your own infrastructure, install Jentic One from its GitHub repo.