Agent Skills give AI coding agents — for example, Claude Code — built-in, validated knowledge of how AIDB, PGAA, and PGFS actually work. Each skill is maintained by EDB alongside the product it covers: its SQL functions, configuration, and concepts, kept in sync as the product changes, in a form the agent reads directly rather than one it has to piece together on its own.
Why install EDB's Agent Skills
Accurate to the product you're running: each skill reflects AIDB, PGAA, or PGFS's actual SQL functions, configuration, and concepts, maintained by EDB alongside the product itself.
Specific, not generic: skill content is scoped to EDB Postgres AI products, so the agent works from the product's real surface instead of general Postgres knowledge that may not apply.
Stays current: as AIDB, PGAA, or PGFS change, the skill changes with them, so it doesn't go stale the way a scraped blog post or an older training snapshot would.
What Agent Skills cover
Skills are available for:
| Product | What it does |
|---|---|
| AIDB | Brings native AI capabilities — embeddings, RAG pipelines, and in-database agents — into Postgres. |
| Postgres Analytics Accelerator (PGAA) | Analytics in Postgres over Apache Iceberg, Delta Lake, and Parquet tables. |
| Postgres File System (PGFS) | Companion extension for AIDB and PGAA that provides access to external storage. |
Getting the skills
The skills are maintained in the edb-agent-skills GitHub repository.
Installing a skill
Decide which product's skill you want to install, or install all of them by not selecting a specific directory.
Ask your AI agent to install the skill, for example:
Install the skills from https://github.com/EnterpriseDB/edb-agent-skills.Follow any instructions the agent gives you to complete the installation.
Once installed, a skill runs as part of your agent's local session — there's no separate service to deploy or account to configure. The agent loads the skill's content itself when it decides the skill is relevant to your request.
Using the skills
Once a skill is installed, ask your agent to help with tasks like:
Discovery and solution design — for example,
How can I build a back end for a RAG application with EDB AIDB?SQL generation, or asking the agent to run SQL directly
Installing and configuring the product — for example, "Register a new LLM for use with AIDB; I want to use OpenAI GPT-5"
Tuning, experimentation, and optimization — for example, "I want to try different chunking strategies for my RAG vector embeddings"
Debugging — for example, "My RAG knowledge base 'products' returns poor results; what could the reason be?"
Coding — for example, "Integrate the local Python demo app with PG and the AIDB extension so I can add semantic search to the UI"
Example
This partial example, from a Claude Code session with the AIDB skill installed, shows the agent answering a discovery question:
❯ what can I do with the AIDB extension? ⏺ Skill(aidb) ⎿ Successfully loaded skill ⏺ Everything AIDB does is a SQL call in the aidb schema — no CLI, no SDK. Here's the capability map, checked against your instance on port 28818. The six building blocks ┌─────────────────────┬──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐ │ Block │ What you get │ ├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ Standalone │ One-off AI inside a query: chunk_text, parse_html, parse_pdf, pdf_to_image, perform_ocr, summarize_text, encode_text, generate_text, rerank_text │ │ functions │ │ ├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤ │ Models │ Register once by name, reuse everywhere. Local (in-process) or remote (OpenAI, Anthropic, Gemini, NIM, OpenRouter) │ ...
Supported AI platforms
- Claude Code