FTS5
PulseAugur coverage of FTS5 — every cluster mentioning FTS5 across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
Developer builds aiZee to govern AI coding agent actions
A developer has created aiZee, a local control plane designed to act as a governance layer for AI coding agents. This system enforces policies on agent actions, preventing unwanted behaviors like committing sensitive in…
-
FTS5 trigger bug silently broke Soma agent's memory updates
A bug in the FTS5 trigger used for updating facts in SQLite silently prevented any updates for two weeks, impacting the Soma agent's memory functionality. The issue stemmed from an incompatibility between the trigram to…
-
Coding agents' memory tested: skillmem shows promise, bugs found
A developer tested the memory capabilities of coding agents, finding that without a dedicated memory system, agents failed to recall information across sessions. When using a system called skillmem, which integrates wit…
-
New tool atmo-mem gives coding agents persistent memory
A new tool called atmo-mem has been released, designed to provide persistent memory for coding agents like Claude Code, Cursor, and Cline. This tool allows agents to save and recall information across sessions, addressi…
-
Developer builds free API for Romania's company registry data
A developer has created a free API for Romania's company registry, which previously lacked programmatic access. The solution involves processing a large CSV dump into a searchable SQLite database with a custom full-text…
-
Engram's persistent memory system debugged for coding agents
The author details their experience using Engram, a local persistent memory system for coding agents, during a short break. They found the architecture straightforward, utilizing Go, SQLite with FTS5, and an MCP server,…
-
DocuSearch system enhances enterprise search with hybrid RAG and KG expansion
A new research paper details DocuSearch, an enterprise document search system designed to improve accuracy and grounding for complex queries. DocuSearch integrates semantic search with BGE-Large embeddings, BM25 full-te…
-
PipeWise uses LLMs to turn plumbing subreddit posts into content
The PipeWise content engine transforms a plumbing subreddit's raw posts into valuable blog content. It scrapes posts, enriches them with a local Qwen2.5 model for tagging, and stores them in SQLite. The system then clus…
-
AI agents can use simpler knowledge bases instead of complex RAG
An AI agent does not always require a complex Retrieval-Augmented Generation (RAG) pipeline to access organizational knowledge. For many common tasks, such as finding a policy or updating a document, a simpler approach …
-
New scrydb library enables semantic and hybrid search in SQLite
A new Python library called scrydb has been developed, enabling users to perform lexical, semantic, and hybrid search directly within SQLite databases. It integrates SQLite's FTS5 extension for traditional keyword searc…
-
Local RAG development prioritizes retrieval over infrastructure
This article advocates for a "local first" approach to developing retrieval-augmented generation (RAG) systems, emphasizing the importance of starting with retrieval rather than complex infrastructure. The author sugges…
-
Developer builds local AI retrieval system to cut token costs
A developer has created a local-first retrieval system to efficiently answer questions from a large knowledge base of markdown files. The system uses SQLite with FTS5 for fast keyword searching and exact filtering, comb…
-
AI models struggle with confidentiality; new fast memory tech unveiled
A new project called MCP Memory has been released, utilizing Google's Open Knowledge Foundation (OKF) and SQLite's FTS5 for fast agent memory. Separately, a user shared an experience where Claude Code, an AI model, fail…
-
TopVideoHub uses Claude API for multilingual video metadata moderation
TopVideoHub has implemented a new content moderation system using the Claude API to handle multilingual video metadata, particularly for Asian languages like Japanese and Korean. The previous regex-based blocklist appro…
-
RAG system debugging reveals noise, retrieval failures, and model knowledge leaks
A developer detailed a challenging debugging process involving a retrieval-augmented generation (RAG) system. Initially, the system appeared to regress in performance, but a meticulous byte-by-byte comparison of request…
-
New tool memcp enables AI coding agents to share memory
A new open-source tool called memcp has been developed to address the problem of AI coding agents starting from scratch with each new session. Memcp captures session logs from different AI coding agents, such as Claude …
-
Patent database scales to 5.36M records with SQLite/FTS5 optimizations
A patent lawyer and hobby coder shared insights on scaling a patent database from 3.5 million to 5.36 million records using SQLite and FTS5. Key learnings include the importance of running ANALYZE after bulk loading, th…
-
SQLite FTS5 outperforms vector search for coding agent memory
A developer found that SQLite's FTS5 full-text search was a more effective tool than vector search for their coding agent's memory. While vector search excels at semantic similarity for prose, FTS5 is better suited for …
-
Svarna workbench offers open access to Modern Greek language corpora
A new open-source workbench called Svarna has been developed for Modern Greek language corpora. This web-based platform integrates five diverse databases, totaling over 507 million words and 29 million sentences, to add…
-
Kiro AI agent gains persistent memory with new open-source tool
A new open-source tool called Kiro-Ception has been developed to give the Kiro AI agent persistent memory across all user sessions and devices. This tool indexes conversation history locally, utilizing a hybrid search a…