AI Data Directory

Open-source web crawler designed for LLMs and AI agents with structured extraction and browser automation.

Open-source local-first cognitive memory system implementing AGM-compatible belief revision that automatically re-evaluates downstream beliefs when facts change, with SHA-256 hash chain for data integrity.

A deterministic SQL semantic inspector that catches silently-wrong AI-generated queries — double-counting, bad joins, exposed PII — in about 0.1 ms before they run. Works as a CI gate, an MCP server, or a library.

Adaptive Recall is a hosted memory system for AI applications that goes far beyond simple vector search. It stores, recalls, and manages long-term memory for agents and apps over MCP or a plain REST API, and — unlike a static embeddings store — it actively learns. Four retrieval strategies run in parallel (vector similarity, temporal recency, full-text keyword, and knowledge-graph traversal), and the system learns which to prioritize for each query type. Results are ranked with ACT-R cognitive scoring from 30 years of cognitive-science research, factoring in recency, access frequency, entity connections, and validated confidence. A knowledge graph is built automatically from stored memories, memories move through a confidence-based lifecycle and fade when unused, and an ML pipeline trains on your usage patterns — validating every parameter change against real query history before adopting it. A simple eight-tool API (store, recall, update, forget, graph, status, snapshot, feedback) covers everything, with Bearer-token auth and JSON in/out. Free, Starter, Pro, and Business plans are available.

Postgres extensions for retrieval-augmented generation: graph search, multi-path hybrid retrieval, and token-budgeted context assembly, self-hosted or managed.

TamedTable is an AI ETL tool you drive with natural language. Load a CSV, JSONL, Parquet, or Arrow file, type 'normalize phone numbers' or 'drop duplicate emails', and the LLM writes a JSON spec that transforms the data — with a 96.7% label-match benchmark at about $0.15 per 1,000 rows. It cleans, enriches, classifies, validates, and translates; every change saves as a replayable recipe or exportable Python script. Source-available, runs on your own API keys (BYOK).

An open-source RAG framework (Apache 2.0) that hits research-SOTA parity on multi-hop QA benchmarks using only commodity LLM APIs — no GPU, no training, no graph rebuild.

ParseHawk is a fully local document AI processing toolkit — no data leaves your machine. It ships with an API server, CLI, and Web UI, making it easy to integrate into existing workflows or use standalone for document parsing, chunking, OCR, and Q&A over documents.