Godantic

Pydantic for Go - JSON schema generation, LLM response validation, partial JSON streaming. Catch hallucinated types before they crash your app. Works with OpenAI, Anthropic, Gemini, Gin REST APIs (with OpenAPI generation), and any system that consumes JSON Schema. Inspired by Python's Pydantic and FastAPI.

func (u *User) FieldEmail() godantic.FieldOptions[string] {
    return godantic.Field(
        godantic.Required[string](),
        godantic.Email(),
        godantic.Description[string]("Primary contact email"),
    )
}

Key Features:

• LLM-ready schemas: Generate flattened JSON Schema for OpenAI, Anthropic, Gemini structured outputs
• Streaming partial JSON: Parse incomplete JSON as it streams from LLM APIs for real-time UI updates
• Runtime validation: Catch hallucinated types, missing fields, and invalid enums before they crash your app
• Union type support: Full anyOf/oneOf schemas with discriminators
• No struct tags: Uses Field{Name}() methods—works with your IDE, tests, and debugger
• Gin integration: Automatic OpenAPI generation and interactive docs

go get github.com/deepankarm/godantic

LLM Structured Output

Godantic solves a common problem: LLMs sometimes return invalid data (wrong types, missing fields, out-of-range values). Define your schema once, use it to guide LLM responses AND validate them.

Basic Workflow

import (
    "github.com/deepankarm/godantic"
    "github.com/deepankarm/godantic/schema"
)

// 1. Define your response structure
type TaskList struct {
    Tasks []Task
}

type Task struct {
    Title    string
    Priority string
    Done     bool
}

func (t *TaskList) FieldTasks() godantic.FieldOptions[[]Task] {
    return godantic.Field(
        godantic.Required[[]Task](),
        godantic.MinItems[Task](1),
    )
}

func (t *Task) FieldPriority() godantic.FieldOptions[string] {
    return godantic.Field(
        godantic.OneOf("low", "medium", "high"),
    )
}

// 2. Generate flattened schema for LLM APIs
schemaGen := schema.NewGenerator[TaskList]()
flatSchema, _ := schemaGen.GenerateFlattened()

// 3. Send schema to your LLM provider (OpenAI, Gemini, Claude)
jsonData := callLLM(prompt, flatSchema)

// 4. Validate response before using it
validator := godantic.NewValidator[TaskList]()
result, errs := validator.Unmarshal(jsonData)
if len(errs) > 0 {
    // LLM returned invalid data - handle or retry
}

Streaming Partial JSON

Parse incomplete JSON as it streams from LLM APIs. Essential for real-time UI updates during long-running generation.

// Create streaming parser
parser := godantic.NewStreamParser[Response]()

// Feed chunks as they arrive from LLM API
for chunk := range llmStream {
    result, state, _ := parser.Feed(chunk)
    
    if state.IsComplete {
        // Full response received and validated
        fmt.Println("Complete:", result)
    } else {
        // Show partial data in real-time
        // state.WaitingFor() tells you which fields are still incomplete
        fmt.Printf("Streaming... waiting for: %v\n", state.WaitingFor())
    }
}

How it works:

• Repairs incomplete JSON (closes unclosed strings, arrays, objects)
• Tracks which fields are still being streamed via state.WaitingFor()
• Skips validation for incomplete fields
• Applies defaults automatically

See examples/llm-partialjson-streaming/ for a complete working example with Gemini streaming.

Provider Examples

See complete working examples:

• examples/openai-structured-output/ - OpenAI structured output API
• examples/gemini-structured-output/ - Google Gemini with enums and unions

Quick Start

import "github.com/deepankarm/godantic"

type User struct {
    Email    string
    Age      int
    Username string
}

// Define validation rules
func (u *User) FieldEmail() godantic.FieldOptions[string] {
    return godantic.Field(
        godantic.Required[string](),
        godantic.Email(),
    )
}

func (u *User) FieldAge() godantic.FieldOptions[int] {
    return godantic.Field(
        godantic.Min(0),
        godantic.Max(130),
    )
}

func (u *User) FieldUsername() godantic.FieldOptions[string] {
    return godantic.Field(
        godantic.Required[string](),
        godantic.MinLen(3),
        godantic.MaxLen(20),
    )
}

// Validate
validator := godantic.NewValidator[User]()
errs := validator.Validate(&user)

Features

Type-Safe Constraints

// Numeric constraints
godantic.Field(
    godantic.Min(0),
    godantic.Max(100),
    godantic.MultipleOf(5),
)

// String constraints
godantic.Field(
    godantic.MinLen(3),
    godantic.MaxLen(20),
    godantic.Regex("^[a-zA-Z0-9]+$"),
    godantic.Email(),
    godantic.URL(),
)

// Enum validation
godantic.Field(
    godantic.OneOf("active", "inactive", "pending"),
)

Custom Validation

func (u *User) FieldPassword() godantic.FieldOptions[string] {
    return godantic.Field(
        godantic.Required[string](),
        godantic.MinLen(8),
        godantic.Validate(func(password string) error {
            // Your custom logic
            if !hasUppercase(password) {
                return fmt.Errorf("must contain uppercase")
            }
            return nil
        }),
    )
}

Union Types

By design, Go doesn't have native union types. However, when building systems that interact with external APIs, LLMs, or generate OpenAPI schemas, you often need to express "this field can be one of several types" in JSON Schema.

Godantic provides runtime validation and JSON Schema generation for union-like patterns using Go's existing type system (any and interfaces).

1. Simple Unions (anyOf) - For any fields

Runtime validation + schema generation for fields that accept multiple types:

type APIRequest struct {
    Query any  // Accept string OR array of inputs
}

func (r *APIRequest) FieldQuery() godantic.FieldOptions[any] {
    return godantic.Field(
        godantic.Union[any]("string", []TextInput{}, []ImageInput{}),
        godantic.Description[any]("Text query or structured inputs"),
    )
}

Generated JSON Schema:

{
  "Query": {
    "anyOf": [
      {"type": "string"},
      {"type": "array", "items": {"$ref": "#/$defs/TextInput"}},
      {"type": "array", "items": {"$ref": "#/$defs/ImageInput"}}
    ]
  }
}

2. Discriminated Unions (oneOf) - For interface fields

Runtime validation + schema generation with discriminator field (OpenAPI 3.1, LLM tool calls):

type ResponseType interface {
    GetStatus() string
}

type SuccessResponse struct {
    Status string            // Must be "success"
    Data   map[string]string
}

type ErrorResponse struct {
    Status  string  // Must be "error"
    Message string
    Code    int
}

type APIResult struct {
    Response ResponseType  // Compile-time type safety via interface
}

func (a *APIResult) FieldResponse() godantic.FieldOptions[ResponseType] {
    return godantic.Field(
        godantic.DiscriminatedUnion[ResponseType]("Status", map[string]any{
            "success": SuccessResponse{},
            "error":   ErrorResponse{},
        }),
    )
}

Generated JSON Schema:

{
  "Response": {
    "oneOf": [
      {"$ref": "#/$defs/SuccessResponse"},
      {"$ref": "#/$defs/ErrorResponse"}
    ],
    "discriminator": {
      "propertyName": "Status",
      "mapping": {
        "success": "#/$defs/SuccessResponse",
        "error": "#/$defs/ErrorResponse"
      }
    }
  }
}

What validation provides:

• Validates Status field is one of ["success", "error"]
• Catches typos: Status: "succes" → validation error
• Prevents invalid states: ErrorResponse{Status: "success"} → validation error
• Works with Go's type system (interfaces for compile-time safety)
• Generates proper JSON Schema for external tools

Real-world benefits:

• LLM integrations: Send proper schemas for structured outputs (OpenAI, Anthropic, Gemini)
• OpenAPI generation: Create API specs that generate type-safe clients
• API validation: Catch data inconsistencies at runtime
• Interoperability: Work with systems that expect union types (TypeScript, Pydantic)

Validating interfaces

Automatic type routing for interface types based on discriminator fields. Useful for polymorphic API requests, payment methods, or any domain where you need runtime type selection.

type PaymentType string

const (
    PaymentTypeCreditCard   PaymentType = "credit_card"
    PaymentTypePayPal       PaymentType = "paypal"
    PaymentTypeBankTransfer PaymentType = "bank_transfer"
)

type PaymentMethod interface {
    GetType() PaymentType
}

type CreditCardPayment struct {
    Type       PaymentType `json:"type"`
    CardNumber string      `json:"card_number"`
}

type PayPalPayment struct {
    Type  PaymentType `json:"type"`
    Email string      `json:"email"`
}

type BankTransferPayment struct {
    Type        PaymentType `json:"type"`
    AccountName string      `json:"account_name"`
}

// Create validator with discriminator
validator := godantic.NewValidator[PaymentMethod](
    godantic.WithDiscriminatorTyped("type", map[PaymentType]any{
        PaymentTypeCreditCard:   CreditCardPayment{},
        PaymentTypePayPal:       PayPalPayment{},
        PaymentTypeBankTransfer: BankTransferPayment{},
    }),
)

// Automatically routes to correct type based on "type" field
payment, errs := validator.Unmarshal(jsonData)

// Type switch for handling
switch p := (*payment).(type) {
case CreditCardPayment:
    fmt.Printf("Credit card: %s\n", p.CardNumber)
case PayPalPayment:
    fmt.Printf("PayPal: %s\n", p.Email)
case BankTransferPayment:
    fmt.Printf("Bank: %s\n", p.AccountName)
}

How it works:

1. Examines discriminator field ("type") to determine concrete type
2. Unmarshals JSON into the appropriate struct (e.g., CreditCardPayment)
3. Validates all fields according to their Field methods
4. Applies default values if defined
5. Returns as interface type with proper concrete value

Key benefits:

• No manual discriminator routing code required
• Compile-time type safety via Go interfaces
• All field validation rules applied automatically
• Clean separation of validation logic from business logic

See examples/payment-methods/ for a complete working example.

JSON Schema Generation

Generate JSON Schema without struct tags:

import "github.com/deepankarm/godantic/schema"

sg := schema.NewGenerator[User]()
s, err := sg.Generate()

// Or as JSON string
jsonSchema, err := sg.GenerateJSON()

All validation constraints (min, max, pattern, etc.) are automatically included in the schema.

JSON Marshal/Unmarshal with Validation

Godantic provides convenient methods for working with JSON that automatically apply defaults and validate:

Unmarshal - JSON → Struct (with validation)

Unmarshals JSON, applies defaults, and validates in one step:

validator := godantic.NewValidator[User]()

// One-liner: unmarshal + defaults + validate
user, errs := validator.Unmarshal(jsonData)
if len(errs) > 0 {
    // Handle validation errors
}
// user is ready to use with all defaults applied

Marshal - Struct → JSON (with validation)

Validates, applies defaults, and marshals to JSON in one step:

validator := godantic.NewValidator[User]()

// One-liner: validate + defaults + marshal
jsonData, errs := validator.Marshal(&user)
if len(errs) > 0 {
    // Handle validation errors
}
// jsonData is valid JSON with all defaults

Lifecycle Hooks

Godantic provides hooks to transform data at different stages of validation and serialization:

BeforeValidate - Transform JSON before validation

Modify the raw JSON map before unmarshaling and validation (useful for backward compatibility):

func (u *User) BeforeValidate(raw map[string]any) error {
    // Normalize legacy format
    if legacyField, ok := raw["old_field"]; ok {
        raw["new_field"] = legacyField
        delete(raw, "old_field")
    }
    return nil
}

AfterValidate - Transform struct after validation

Modify the struct after successful validation (useful for computed fields):

func (u *User) AfterValidate() error {
    u.DisplayName = strings.ToLower(u.Username)
    return nil
}

BeforeSerialize - Transform struct before marshaling

Modify the struct before marshaling to JSON:

func (u *User) BeforeSerialize() error {
    u.UpdatedAt = time.Now()
    return nil
}

AfterSerialize - Transform JSON after marshaling

Modify the JSON bytes after marshaling (useful for wrapping responses):

func (u *User) AfterSerialize(data []byte) ([]byte, error) {
    // Wrap in envelope
    envelope := map[string]any{
        "data": json.RawMessage(data),
        "version": "1.0",
    }
    return json.Marshal(envelope)
}

Execution order:

• Unmarshal (JSON → Struct): BeforeValidate → unmarshal → validate → defaults → AfterValidate
• Marshal (Struct → JSON): BeforeSerialize → validate → defaults → marshal → AfterSerialize

Complex Structures

Works with embedded structs, nested structs, pointers, slices, and maps:

type Company struct {
    Name     string
    Address  Address        // Nested struct
    Contacts []Contact      // Slice of structs
    Settings map[string]int // Map
}

Gin Integration (gingodantic)

FastAPI experience with Gin. Automatic OpenAPI generation, request validation, and interactive docs—define your types once, get everything else for free.

import "github.com/deepankarm/godantic/pkg/gingodantic"

type CreateUserRequest struct {
    Name  string `json:"name"`
    Email string `json:"email"`
}

func (c *CreateUserRequest) FieldName() godantic.FieldOptions[string] {
    return godantic.Field(godantic.Required[string](), godantic.MinLen(2))
}

func (c *CreateUserRequest) FieldEmail() godantic.FieldOptions[string] {
    return godantic.Field(godantic.Required[string](), godantic.Email())
}

// Setup
router := gin.Default()
api := gingodantic.New("User API", "1.0.0")

// Register endpoint with schema
router.POST("/users",
    api.OpenAPISchema("POST", "/users",
        gingodantic.WithSummary("Create user"),
        gingodantic.WithRequest[CreateUserRequest](),  // Enables validation
        gingodantic.WithResponse[UserResponse](201, "Created"),
    ),
    func(c *gin.Context) {
        req, _ := gingodantic.GetValidated[CreateUserRequest](c)
        // req is validated and ready to use
    },
)

// Serve OpenAPI spec and documentation UIs
router.GET("/openapi.json", api.OpenAPIHandler())
router.GET("/docs", gingodantic.SwaggerUI("/openapi.json"))
router.GET("/redoc", gingodantic.ReDoc("/openapi.json"))  // Alternative UI

Features:

• Automatic validation: Request bodies, query params, path params, headers, and cookies
• OpenAPI 3.0.3 generation: Complete spec with all parameter types and constraints
• Type-safe helpers: GetValidated[T](), GetValidatedQuery[T](), GetValidatedPath[T](), etc.
• Validation by default: Enabled automatically when request types are specified
• Documentation UIs: Built-in Swagger UI and ReDoc handlers
• Zero boilerplate: No manual schema writing or validation middleware

Parameter types supported:

gingodantic.WithRequest[T]()        // Request body
gingodantic.WithQueryParams[T]()    // Query parameters
gingodantic.WithPathParams[T]()     // Path parameters (:id)
gingodantic.WithHeaderParams[T]()   // Request headers
gingodantic.WithCookieParams[T]()   // Cookies
gingodantic.WithResponse[T](code)   // Response schemas

All godantic constraints (min, max, regex, email, etc.) are automatically included in the OpenAPI spec.

See examples/gin-api/ for a complete working API with all parameter types.

Available Constraints

godantic.Required[T]()              // required field

// numeric constraints
godantic.Min(value)                 // value >= min
godantic.Max(value)                 // value <= max
godantic.ExclusiveMin(value)        // value > min
godantic.ExclusiveMax(value)        // value < max
godantic.MultipleOf(value)          // value is multiple of

// string constraints
godantic.MinLen(length)             // minimum length
godantic.MaxLen(length)             // maximum length
godantic.Regex(pattern)             // regex pattern match
godantic.Email()                    // email format
godantic.URL()                      // URL format
godantic.ContentEncoding(encoding)  // e.g., "base64"
godantic.ContentMediaType(type)     // e.g., "application/json"

// array/slice constraints
godantic.MinItems[T](count)         // minimum number of items
godantic.MaxItems[T](count)         // maximum number of items
godantic.UniqueItems[T]()           // all items must be unique

// map/object constraints
godantic.MinProperties(count)       // minimum properties
godantic.MaxProperties(count)       // maximum properties

// union constraints
godantic.Union[T](type1, type2, ...) // any of the types
godantic.DiscriminatedUnion[T](discriminator, map[string]any{
    "type1": type1{},
    "type2": type2{},
    ...
}) // one of the types based on discriminator

// value constraints
godantic.OneOf(value1, value2, ...) // enum - one of allowed values
godantic.Const(value)               // must equal exactly this value
godantic.Default(value)             // default value (schema only)

// schema metadata
godantic.Description[T](text)       // field description
godantic.Example(value)             // example value
godantic.Title[T](text)             // field title
godantic.Format[T](format)          // format hint (e.g., "date-time")
godantic.ReadOnly[T]()              // read-only field
godantic.WriteOnly[T]()             // write-only field
godantic.Deprecated[T]()            // deprecated field

Custom Validation

godantic.Validate(func(val T) error {
    // Your custom validation logic
    return nil
})

How it works

1. Define Field{FieldName}() methods that return FieldOptions[T]
2. Use Field() to compose validation constraints
3. Create a Validator[T] and call Validate()
4. Generate JSON Schema with schema.NewGenerator[T]()

Zero values (empty string, 0, nil) are treated as "not set" for required field checks.

Testing

go test ./... -v -cover

---

Examples

Check out examples/ for complete working examples:

• openai-structured-output/ - Using godantic with OpenAI's structured output API to extract meeting summaries
• gemini-structured-output/ - Using godantic with Google Gemini to parse task lists with enums and unions
• llm-partialjson-streaming/ - Streaming partial JSON during long-running generation using Gemini
• payment-methods/ - Validating payment requests interfaces using discriminated unions
• gin-api/ - Complete Gin REST API with automatic OpenAPI generation and Swagger UI