Files and databases.
One interface.
A consistent data context lets your agent reason across different sources.
Explore data sources
Skip to featuresFrom Raw Data to
An answer with
The right chart.
Data Security.
Accuracy you
Ready for your agent.
Trust the Analysis,
Bring Analytics
Built on AVA.
yvonneyx/ava-notebook
HellowVirgil/dsh-antv-ava
yvonneyx/ava-engine-duckdb-wasmlyw405/ava-observability 
antvis/AVA
antvis/AVA #1035 antvis/AVA #1035 How AVA Works.
Your agent’s next skill?
NEWIntroducing AVA 4.01.6k Visual Analytics.
Visual Analytics.
Open Source, Built for Agents.
An open-source framework for AI agents to explore, analyze, and visualize data.
From Raw Data to
Answers You Can Inspect.
Files and databases.
An answer with
an evidence trail.
Executed queries and their results give each answer a checkable basis.
Explore trusted analysis
The right chart.
For your question.
The question and the shape of the data guide the chart choice.
Explore intelligent visualization
Data Security.
Run AVA Locally.
You control where analysis runs and which model endpoint it uses.
Explore local enginesAccuracy you
can inspect.
Accuracy is measured against reference answers under published test conditions.
Explore published evaluations
Ready for your agent.
Built to reuse.
Your agent provides the reasoning; AVA executes the data workflow.
Get started with AVA
Trust the Analysis,
Not Just the Answer.
Measured on public benchmarks. Built for answers you can verify.
Context · DataBench · 1,789 questions
93.01%
answer accuracy with Semantic
model: GLM-5.1 · DuckDB · Direct
Raw data83.51%
Schema + Profile84.18%(+0.67 pp)
Semantic★93.01%(+9.50 pp)
Grounded in your data.
Your schema and data context give the analysis a factual starting point.
Based on real computation.
Answers are built from executed queries and returned results.
Open to inspection.
Follow the queries and results behind the answer, and check them yourself.
Bring Analytics
to Your Agent.
Choose an Agent Skill, CLI, or TypeScript SDK.
bash
# install the skills for your agents
$ npx skills add antvis/AVA
# or add the marketplace in Claude Code
$ /plugin marketplace add antvis/AVA
# then, in a new agent session:
"Use the AVA skill to analyze /absolute/path/sales.csv,
compare sales by region, and generate a chart."Built on AVA.
Shared by the Community.
Projects, extensions, and ideas from the AVA community.
AVA Notebook
CommunityRevisit the SQL, results, and evidence behind your analysis in an editable notebook.
DeepSeek Harness
CommunityBring AVA v4 queries, streaming narratives, and AntV charts into DSH conversations.
DuckDB-Wasm
CommunityRun SQL on CSV data in a browser worker, then analyze and visualize with AVA.
ava-observability
CommunityTrace analysis steps, timings, and SQL attempts through AVA’s public events.
AVA Agent Skill
OfficialExplore data and generate charts directly from the AI agent you already use.
AVA Feishu Bot
CommunityA proposed chat bot for querying files and returning answers and charts in Feishu.
AVA Feishu Plugin
CommunityA proposed Bitable plugin for analyzing and visualizing data inside Feishu.
How AVA Works.
What You Need to Start.
What is AVA?
AVA is an open-source visual analytics framework from AntV. It gives AI agents the tools to connect to data, understand its structure, run analysis, and turn results into charts.
How can I get started?
Try the AI playground, add the Agent Skill to your agent, or follow the TypeScript SDK quickstart. You can also use AVA from the command line.
Which data sources can I use?
AVA supports CSV, JSON, Excel, Parquet, and text, plus SQLite, MySQL, PostgreSQL, MongoDB, Supabase, ClickHouse, and Feishu Bitable. Available sources and engines depend on your runtime.
Do I need a separate model API key?
The Agent Skill uses your agent’s own reasoning, so it does not need a separate model key. AI-powered analysis through the SDK or CLI requires your model configuration. Direct queries and data profiling can run without an LLM.
Is AVA free and open source?
Yes. AVA is released under the MIT License. You can inspect the source, build with it, and contribute on GitHub. Model providers and hosted infrastructure may have their own costs.
Your agent’s next skill?
Understanding your data.
Connect, analyze, and visualize — with the tools your agent needs.