Agents need bounded workflows
An agent should not guess which schema to call or spend credits without confirmation. It should discover available schemas, inspect maturity, estimate the job, request confirmation, start extraction, poll status, and retrieve artifacts.
PolDex exposes those steps through API, OpenAPI, CLI, MCP, and machine-readable discovery files. That makes the agent flow explicit instead of hidden in a UI.
Evidence is agent context
When agents act on extracted data, they need to understand confidence, evidence, and uncertainty. A raw JSON field is not enough. The agent needs to know whether the field is present, unresolved, conflicting, or unsupported.
That is why FastScript output includes truth states and evidence posture. It gives agents a safe way to continue, stop, escalate, or ask for another document.
The same rails serve humans and agents
The processor, API, CLI, MCP server, webhooks, and exports all wrap the same result. There is no separate agent product with different behavior.
That matters for trust. If an operator reviews an artifact and an agent calls the API, both should see the same schema-controlled truth source.