# PolDex

PolDex is the vertical API for insurance document extraction. Commercial insurance is live first; the system is designed to expand line by line into broader insurance document families.

PolDex is useful for developers, workflow tools, AI agents, brokers, underwriters, compliance teams, procurement teams, and insurance-native service companies that need evidence-backed structured JSON from messy insurance documents.

## Canonical URLs

- Site: https://poldex-web.pages.dev
- Docs: https://poldex-web.pages.dev/docs
- Full LLM context: https://poldex-web.pages.dev/llms-full.txt
- Agent manifest: https://poldex-web.pages.dev/.well-known/poldex-agent.json
- OpenAPI: https://poldex-web.pages.dev/openapi.json
- API base: https://poldex-api.plain-water-cd0f.workers.dev
- MCP Registry: https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.lordolami%2Fpoldex

## What PolDex Extracts

PolDex extracts schema-constrained, evidence-backed JSON from insurance documents. The first live schema family is commercial P&C insurance:

- commercial_gl
- commercial_auto
- workers_comp
- umbrella_excess
- commercial_property
- professional_lines

Common document profiles include COIs, ACORD-style forms, declarations, schedules, endorsements, policy packets, requirements packets, and related insurance evidence documents.

## Agent Interfaces

Agents can call PolDex through:

- REST API: https://poldex-web.pages.dev/openapi.json
- MCP server: `npm install -g @poldex/mcp-server`
- CLI: `npm install -g @poldex/cli`
- Machine-readable manifest: https://poldex-web.pages.dev/.well-known/poldex-agent.json

## Safe Agent Workflow

1. Discover schemas with `GET /v1/schemas`.
2. Estimate pages and credit cost with `POST /v1/batches/estimate`.
3. Ask for confirmation before spending credits.
4. Process files, URLs, or text with `POST /v1/batches`.
5. Poll batch status or listen for connector/webhook events.
6. Download JSON, CSV, XLSX, or ZIP artifacts.

## Safety Rules

- Do not claim PolDex has published 99% accuracy unless a public benchmark run has passed the release gate.
- The 99% benchmark is a release gate and benchmark program, not a current universal accuracy claim.
- Treat evidence pointers and truth states as required output fields when evaluating reliability.
- Batch extraction spends credits only after explicit action.
- CLI extraction requires `--yes`.
- MCP extraction requires `confirm_cost: true`.

## Useful Commands

```bash
curl https://poldex-api.plain-water-cd0f.workers.dev/v1/schemas
curl https://poldex-api.plain-water-cd0f.workers.dev/v1/schemas/commercial_gl
curl https://poldex-web.pages.dev/openapi.json
curl https://poldex-web.pages.dev/.well-known/poldex-agent.json
```