Free Tier Architecture Efficiency

7/12/2023

This post details the architectural changes enabling the removal of the Pinecone waitlist and inactivity policy for free tier users. Key technical contributions include the introduction of a new, more efficient Pinecone architecture and a new free region (`gcp-starter`). This new architecture allows for instant sign-ups, indefinite free index retention, and sub-second index provisioning. It also involves simplifying free indexes by removing features like namespaces, collections, and delete by metadata to optimize for efficiency and ease of use for new developers.

4/26/2023

The free plan's capacity has been adjusted to approximately 100k 1536-dimensional embeddings with metadata, aligning with the usage patterns of 99.1% of free users. Paid users can now retain a single free index as a sandbox after upgrading. The architecture has been simplified, removing the need for users to select pod types, and index provisioning time has been reduced to seconds. Inactive free indexes will be archived after 7 days of inactivity (1 day for ephemeral indexes), and can be recreated from collections.

4/14/2023

This post announces architectural changes to the free tier to support a rapidly growing number of daily signups. Key changes include expanding cloud capacity, implementing a waitlist for new users, and reducing the inactivity period for archiving free indexes from 14 days to 7 days (or 1 day for ephemeral indexes). The post also mentions exploring further architectural changes to increase long-term user support.

3/9/2023

This post details a partial database outage that affected a portion of indexes on the Starter (free) plan due to a bug in an inactivity script. 515 indexes were inadvertently deleted and then fully restored two days later. No indexes on paid plans were affected. The root cause was a bug in a SQL query and an incorrect deployment process for the script. Actions taken include enhanced approval processes for data modification, auditing of access and permissions, enabling bucket versioning, and prioritizing the development of a new free plan architecture, more educational materials, integration tests for cron jobs, mirroring staging environments to production, enabling soft deletes, and designing a data export feature.

10/4/2021

This post announces the general availability of Pinecone 2.0 with new features and significantly reduced pricing, including a free tier. The free tier offers a single P1 pod capable of searching up to a million vectors with ~100ms latency or 100K vectors in ~20ms. The Standard plan allows selection of P1 or S1 pods (storage-optimized) and replicas, with pricing based on pod usage per hour. The Dedicated plan offers custom deployments in dedicated GCP or AWS environments for very large workloads or specific compliance needs.