Let AI explore. Not improvise.
AI builds a constrained RulePlan from list and detail samples. The server validates and compiles it into extraction rules, never arbitrary model-generated code.
OPEN SOURCE · SELF-HOSTED · TRUSTED WEB DATA
Web data. With proof.
AI understands the page. You approve the rule. Every run is deterministic. Every item has a story.
01 / THE PHILOSOPHY
Websites change. Models are probabilistic. Your data pipeline should not be a black box. Extrio separates discovery from execution, with an explicit boundary at every step.
AI builds a constrained RulePlan from list and detail samples. The server validates and compiles it into extraction rules, never arbitrary model-generated code.
Review fields, selectors, and sample validation together. A human approves publication. The immutable rule version is bound to its review with an Ed25519 signature.
Runs pin a rule and data contract, recording quality decisions, checkpoints, and lineage. Site changes produce new candidates for review, never silent rule mutations.
Not another crawler. A reviewable, repeatable, traceable way to work with web data.
THE EXTRIO WAY02 / THE WORKSPACE
A desktop workspace built for data operations. Rules, runs, and results stay connected, so the important questions do not disappear into logs.

Collected results and unconfirmed source coverage are reported separately and honestly.
FIG. 02REAL LOCAL REVIEW CAPTURES · CHINESE CONSOLE · NOT A LIVE DEMO OR BENCHMARK
For teams that need to explain their data
03 / UNDER THE HOOD
The control plane accepts intent. Independent workers do the work. AI helps generate rules; production runs follow the reviewed contract.
Auth · Contracts · Durable commands · Operation state
/api/v1Crawl4AIPublic or authorized list/detail sources
RulePlan → GatherSpecModel plans. Server validates.
RuleVersionImmutable, after human approval
Exploration and rule compilation use a model. Human approval is a mandatory publication boundary.
Current single-instance topology. Distributed high availability and multi-tenancy are not current release claims.
Explore the architectureMCP exposes Extrio as a governed data tool. Agents can queue collection requests for review, trigger published rules, and query results with lineage. They cannot publish rules.
Explore MCP integrationcreate_collectionhuman reviewtrigger_runpublished rulesquery_itemswith lineagepublish_rulehuman only04 / BUILT IN THE OPEN
Read the code, the contracts, and the decisions. We would rather publish our boundaries and verification evidence than make promises you cannot reproduce.
AI rule generation and repair, human publication, deterministic collection, evidence export, Webhook, and MCP.
Cross-site compatibility, browser-rendered sources, and drift recovery. Current evidence is primarily repository tests and local reviews.
Focused on sustained operation and experience acceptance for real teams. Hosted evaluation, public benchmarks, and multi-tenant hardening remain on the roadmap.
05 / YOUR DATA. YOUR INFRASTRUCTURE.
Clone the source. Run it on your infrastructure. No hosted account required. No opaque service between you and your pipeline.
git clone https://github.com/iiwish/extrio.gitcd extriodocker compose up --buildRequires Git and Docker Compose. Create your instance administrator at 127.0.0.1:8080. Read the security and self-hosting guides before exposing the service.