Intelligent Internet open-sources Evoke PostgreSQL AI search plugin
BIBIBI
AT A GLANCE
The report says that Intelligent Internet has open-sourced Evoke, a CPU search plugin that runs on a standard PostgreSQL.
Article
Open source AI Search plugin Evoke: 30000000 parameters, recall approaching that of the 600000000 model
Beating AI News flash, Stability AI co-founder Emad Mostaque's new company Intelligent Internet has open-sourced Evoke. This is a PostgreSQL database search plugin with approximately 30300000 parameters and a AI model built in. After installation, developers can search documents and knowledge bases directly in the database. It can match keywords and find content with similar meanings.
Evoke is based on IBM’s Granite-Embedding-30M-Sparse model. It generates a set of weighted related terms for documents, so it may find corresponding content even when the original text does not contain the search term. Keywords and semantic information share a single search index, eliminating the need to maintain a separate vector index. The model runs directly on a standard CPU and does not require GPU or sending data to an external AI service.
English retrieval tests previously published by the official team show that Evoke achieves a 100 recall rate for relevant documents within the top 66.7%results, higher than the 56.3%achieved by traditional keyword search and approaching the 600000000 of a model with 67.1%parameters.
Evoke's code and model are both open-sourced under the Apache-2.0 protocol, making it suitable for scenarios such as internal enterprise document retrieval and AI Agent knowledge bases.
Original link https://m.theblockbeats.info/flash/371001
Key points
01
The report says that Intelligent Internet has open-sourced Evoke.
02
Evoke is a PostgreSQL database search plugin with approximately 30300000 parameters and a AI model built in.
03
Evoke is based on IBM’s Granite-Embedding-30M-Sparse model and can match keywords and content with similar meanings.
04
Keywords and semantic information share a single search index, eliminating the need to maintain a separate vector index.
05
The report says that the model can run directly on a standard CPU and does not require GPU or sending data to an external AI service.
06
Officially published English retrieval tests previously showed that Evoke's recall rate within the top 100 results was 66.7%while traditional keyword search achieved 56.3%，600000000 and the model with 67.1%。
07
The report says that Evoke's code and model are both open-sourced under the Apache-2.0 protocol.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Evoke puts keyword matching and the ability to find similar meanings into the PostgreSQL. According to the report, documents are converted into a set of weighted related terms, so relevant content may be found even when the search term does not appear in the original text; it can also run locally on a standard CPU.
Why it matters to readers
If the reported functionality and test results can be reproduced, developers may be able to add search capabilities for documents and knowledge bases without separately maintaining a vector index or using GPU or sending data to an external AI service.
For beginners, it provides a way to try keyword and semantic search directly in a PostgreSQL; however, the evidence does not specify the installation method, actual performance, or suitable data scale.
Risks and unknowns
The available evidence does not provide Evoke's official repository or model page, so its release status and Apache-2.0 authorization cannot be independently verified.
The dataset, complete methodology, and reproduction results for the English retrieval tests were not provided.
66.7%The results come from the reported official tests and cannot establish performance on enterprise internal documents or other data.
The evidence does not specify ordinary CPU systems' speed, resource consumption, index maintenance cost, or supported data scale.
Related Developments
Loading event timeline…
Related concepts
recall
This term is not in the glossary yet. Browse related concepts in the glossary.
