
AVA
๐พ AI-native Visual Analytics framework build for agents.
The Lens
By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.
Updated Nov 2026
AVA lets your app or AI agent answer questions about data in plain English. Point it at a CSV or a Postgres database, ask which products dropped most last month, and it writes the query, runs it, and returns the numbers plus a chart. It's from AntV, Ant Group's visualization team, and it's MIT licensed.
No server to run. It's a Node.js library, and by default queries run in an embedded DuckDB, so data stays on your machine. ClickHouse, SQL Server and Snowflake engines are optional. You bring the model key. A CLI and an agent skill let a coding agent drive it.
The README calls it PandasAI for JavaScript, and that's fair. What I like: every answer shows the SQL behind it, checked as read-only, so you can verify the math instead of trusting a chatbot. Solo: free, you pay only for tokens. Small teams: free, good for internal tools. Larger teams: free, but test accuracy on your own data first.
The catch: version 4 shipped in November 2026 after more than two years without a release, and it's a different product. Older AVA was a chart recommendation library with a different API, so most tutorials don't apply.
Free vs Self-Hosted vs Paid
fully freeFree: Everything. MIT-licensed library on npm (@antv/ava) with the SDK, CLI and agent skill. No paid tier, no hosted plan.
Self-hosted: Runs inside your own Node.js process, by default against an embedded DuckDB. Optional engines query Supabase, ClickHouse, SQL Server or Snowflake directly, or run Python through an executor you supply.
Paid: Nothing from AntV. The real cost is the LLM you configure: every analysis and chart is model calls billed by your provider, or free if you point it at a local model endpoint.
The library is free; your bill is model tokens, which scale with how many questions people ask and how big your schemas are.
What to do by team size
- Solo
- free: pay only for model tokens
- Small team
- free: embed it in an internal tool
- Medium team
- free: budget for model usage
- Large team
- free: validate accuracy on your own data before rollout
Get tools like this every Wednesday
One featured tool, three on the radar. No fluff.
A low score is not a verdict on quality. Young and niche tools start low by design. How we calculate scores
Trust Signals
License: MIT License
Use freely, including commercially; keep the license notice.
Commercial use: โ Yes
About
- Owner
- AntV AI Visualization Team (Organization)
- Stars
- 1,637
- Forks
- 170
Explore Further
More tools in the directory
rea
Reverse engineer anything with agents, from app behavior down to native binaries.
462.8k โeverything-claude-code
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
304.0k โhermes-agent
The agent that grows with you
263.7k โ