Changelog

Search now handles both technical terms ('max_connections') and natural language ('how many connections does postgres allow'). We rebuilt the search engine to better match how developers actually ask questions.

Ask 100 questions for the cost of 10 individual calls. Perfect for preflight checks, bulk lookups, and MCP tools that need to query multiple topics at once.

Optimized the entire pipeline for batch requests. Whether you're checking 10 or 100 questions, you'll get answers in under a second end-to-end.

Already asked a question before? It costs 90% less. Asking it again in a batch? Even cheaper. Discounts compound so repeated questions cost almost nothing.

Complete coverage across 182 subtopics including configuration, performance tuning, replication, security, and common errors. Every answer researched from official PostgreSQL documentation.

Full coverage of App Router, Server Components, Server Actions, and the latest Next.js 15 features. Researched from official Vercel documentation.

Once you've asked a question, asking it again costs only 10% of a normal request. This rewards you for finding useful answers and encourages exploration.

Native integration with Claude Code, Cursor, Cline, and any MCP-compatible client. Your AI assistant can now query AgentsKB directly without any configuration.

'agentskb lock' creates a snapshot of answers your project depends on. 'agentskb verify' checks that nothing has changed. Perfect for reproducible builds and team consistency.

Scans your codebase, identifies what technologies you're using, and pre-fetches relevant knowledge. Your AI starts each session already knowing what you need.

300 requests/month with no signup required. Sign in with Google to get 1,000/month. We want you to try AgentsKB before committing.

We started with PostgreSQL because it's one of the most-used databases with the most configuration questions AI gets wrong. Every answer researched from official docs.

We'd rather say 'not found' than guess wrong. If we're not confident in an answer, we don't return it. No hallucinations, no outdated information from training data.