DSPy-style XML format with schema-directed field delimiters.
Each field is wrapped in <name> and </name>. Values are rendered from the
declared schema:
- strings, string enums, and string literals are raw, unescaped text
- numbers, booleans, null, and non-string literals are JSON text
- objects with declared keys are nested
<key>elements in declared order - arrays are repeated
<item>elements - string-keyed records are JSON text, because their keys are data
Parsing follows the same schema. It is not a general XML parser: there is no
entity expansion, no DTD processing, no attributes, and no namespaces, so
& and <!DOCTYPE ...> stay literal text. A string field ends at the
first matching closing tag, so string values that contain their own closing
tag are rejected when rendering. Each output field must appear exactly once.
Options:
:max_reply_bytes- reply size limit, default 1 000 000:max_depth- nesting limit for objects and arrays, default 32
Differences from DSPy 3.5: field types use JSON names and JSON Schema; nested values follow the declared Zoi schema rather than Python type annotations; non-string scalars are strict JSON without repair.