Intelligent Internet, the new company founded by Stability AI co-founder Emad Mostaque, has open-sourced Evoke, a PostgreSQL search plugin that runs an AI model with about 30.3 million parameters inside the database. The tool is designed for document and knowledge base search, letting developers match both exact keywords and semantically similar content after installation. Evoke is built on IBM’s Granite-Embedding-30M-Sparse model and generates weighted related terms for documents, allowing relevant results to surface even when the original text does not contain the exact search query. Keyword and semantic retrieval share the same search index, so users do not need to maintain a separate vector index. According to previously released official English retrieval tests, Evoke reached a 66.7% relevant-document retrieval rate in the top 100 results, above traditional keyword search at 56.3% and close to a 600 million-parameter model at 67.1%. The code and model are released under the Apache-2.0 license, with use cases including internal enterprise document retrieval and knowledge bases for AI agents.
Intelligent Internet, the new company launched by Stability AI co-founder Emad Mostaque, has open-sourced Evoke, a PostgreSQL search plugin with an AI model of roughly 30.3 million parameters built in.
After installation, developers can search documents and knowledge bases directly inside the database. The plugin supports both keyword matching and semantic retrieval for similar meaning.
Built on IBM’s Granite-Embedding-30M-Sparse
Evoke is based on IBM’s Granite-Embedding-30M-Sparse model. It generates a set of weighted related terms for documents, which means relevant content can still be found even if the exact search terms do not appear in the original text.
Keyword and semantic signals use the same search index, removing the need to maintain a separate vector index. The model also runs directly on a standard CPU, without a GPU, and does not require sending data to an external AI service.
Official retrieval test results
According to previously released official English retrieval tests, Evoke achieved a 66.7% rate of retrieving relevant documents within the top 100 results. That was higher than traditional keyword search at 56.3% and close to a 600 million-parameter model at 67.1%.
Both the code and the model are open-sourced under the Apache-2.0 license. The project is aimed at use cases such as internal enterprise document retrieval and knowledge bases for AI agents.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
