Pinecone

Model

1. TouchesModel APIs: Touches

   Model APIs, touches. Pinecone hosts several embedding models, such as llama-text-embed-v2, and reranking models, and bills them per million tokens or per 1,000 requests. Generation models stay with a model API.

2. Model gateway: Touches

   Model gateway, touches. Pinecone routes no model traffic. A gateway sits beside it for generation calls.

Build

1. Orchestration: Touches

   Orchestration, touches. LangChain's PineconeVectorStore works with Pinecone indexes, and Pinecone integrates with LlamaIndex. The agent logic stays in the framework.

2. SupportsTool use: Supports

   Tool use, supports. Through the Pinecone MCP server, AI agents can search Pinecone documentation, manage indexes, upsert data and query indexes. Other tools stay with their own servers.

Data

1. PineconeCoreRetrieval: Core

   Retrieval, core. This is Pinecone's primary role: it serves full-text, semantic, sparse and hybrid search from one index, narrows results with metadata filters, reranks them, and keeps each tenant's data in its own namespace.

Evaluate

1. Evaluation: Touches

   Evaluation, touches. Pinecone serves the retrieval results an evaluation measures. Scoring answers stays with an evaluation product.

Operate

1. Observability: Touches

   Observability, touches. Audit logs record user, service account and API actions on the management and control plane on Enterprise. Tracing an application's requests stays with an observability tool.

2. Cloud deployment: Supports

   Cloud deployment, supports. Pinecone runs on fully managed infrastructure in listed AWS, GCP and Azure regions, and Bring Your Own Cloud on Enterprise deploys the data plane in your own cloud account.

3. Guardrails: Touches

   Guardrails, touches. Pinecone is SOC2 Type II certified and HIPAA compliant and encrypts data at rest and in transit, which secures the data store. Prompt and output guardrails stay with a security product.