Enum OutputMode 

#[non_exhaustive]
pub enum OutputMode {
    Auto,
    Tool,
    Native,
    Prompted,
}

Controls how an agent’s output_schema is enforced.

Strictness

Native is the only mode whose output is constrained by the provider — the response is guaranteed to match the schema (on providers that support it). Tool and Prompted are best-effort: the schema is offered to the model (as a tool or in the prompt) but the model is asked, not forced, to honor it, so the agent re-prompts a bounded number of times and otherwise validate the returned JSON before relying on it. The default Auto is provider-aware: for a tool + schema agent it routes to Tool only on providers whose native constraint would suppress tool calls, and keeps guaranteed Native structured output on providers that compose the two (e.g. OpenAI, Anthropic).

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

Auto

Resolve at request time: OutputMode::Tool when the agent has an output_schema and at least one function tool (and the tool choice permits the output-tool call), otherwise OutputMode::Native. This is the default and only changes behavior for the tool + schema case (which is broken under Native on providers whose native constraint suppresses tool calls).

Tool

Register the schema as a synthetic “output tool” the model calls to finalize. No native structured-output constraint is sent, so the model can freely call its other tools first. Best-effort: the model is instructed to call the output tool but is not forced to, so validate the result. (pydantic-ai ToolOutput.)

Native

Use the provider’s native structured output (format/response_format). Constrains every turn, so the output is guaranteed to match the schema, but may suppress tool calls on some providers (e.g. Ollama). (pydantic-ai NativeOutput.)

Prompted

Inject the schema into the system prompt and return the model’s final text verbatim. The caller parses it — the text is not guaranteed to be clean JSON and may include prose or markdown fences, so extract/validate before deserializing. Useful for weak/local models that lack reliable tool calling or native structured output. (pydantic-ai PromptedOutput.)

Trait Implementations

impl Clone for OutputMode

fn clone(&self) -> OutputMode

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for OutputMode

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for OutputMode

fn default() -> OutputMode

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for OutputMode

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Eq for OutputMode

impl PartialEq for OutputMode

fn eq(&self, other: &OutputMode) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl Serialize for OutputMode

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for OutputMode