Experimental Features¶

This page lists features that are available but still experimental.

Details¶

behavior.batch-generation-jobs¶

• Kind: behavior
• Target: [tool.datamodel-codegen.jobs], --job, --all-jobs
• Since: 0.72.3

Named batch jobs are experimental; their configuration schema, batch output, and transactional/watch execution contracts may change.

Define named generation jobs in [tool.datamodel-codegen.jobs] and select them with --job or --all-jobs. Each selected job has its own input and output, and the full selection is validated before generation begins.

behavior.remote-reference-lock¶

• Kind: behavior
• Target: datamodel-codegen.lock, --lockfile, --update-lock, and --locked
• Since: 0.72.3

Remote reference integrity locking is experimental: the lock document schema and request-identity compatibility may evolve, but integrity mismatches remain fail-closed and credentials are never persisted.

The lock stores opaque SHA-256 request-identity digests and SHA-256 body digests, never response bodies or request values directly. Each saved display origin contains only the scheme, host, and explicit port—never a path, query, or request headers.

cli-option.generate-client¶

• Kind: cli-option
• Target: --generate-client httpx2 and the --client-* options (and their [tool.datamodel-codegen] keys and generate() options)
• Since: 0.84.0

HTTPX2 client generation from the CLI is experimental; its options and generated package may change.

--generate-client httpx2 generates the models at --output and a client package at --client-output; --client-package and --client-model-package name their import paths, and the other --client-* options configure the package. Like model options, they can be set in [tool.datamodel-codegen] of pyproject.toml, in its profiles, and in its jobs. --check, --diff-against, --watch, --output-format json, and the uv add and next-step notice on stderr work as with --generate-server. generate() takes them as generate_client and the client_* options, and returns every generated file without an output. It needs Python 3.11 or later, both to run and as --target-python-version, --use-missing-sentinel for pydantic_v2.dataclass models, --target-pydantic-version 2.14 with --use-missing-sentinel, and --use-msgspec-tagged-unions and --use-msgspec-nullable-enum-alias for msgspec.Struct models; one left unset takes that value, as with --generate-server.

cli-option.generate-schema-validators¶

• Kind: cli-option
• Target: --generate-schema-validators
• Since: 0.66.1

Schema-derived runtime validators are experimental and may change as JSON Schema coverage is expanded.

The option currently targets Pydantic v2 BaseModel output and covers selected object-level rules such as patternProperties, required-only oneOf/anyOf/not groups, and simple if/then/else required-property conditions.

cli-option.generate-server¶

• Kind: cli-option
• Target: --generate-server fastapi and the --server-* options (and their [tool.datamodel-codegen] keys and generate() options)
• Since: 0.84.0

FastAPI server generation from the CLI is experimental; its options and generated package may change.

--generate-server fastapi generates the models at --output and a server package at --server-output, whose main.py serves the application with the implementations in the services package beside it; generation writes each missing module of services once, with stubs, and never overwrites one. --server-package and --server-model-package name their import paths, and the other --server-* options configure the package. Like model options, they can be set in [tool.datamodel-codegen] of pyproject.toml, in its profiles, and in its jobs. --check compares without writing, --diff-against compares the files two inputs render, --watch regenerates after input changes, --output-format json uses the model generation, check, and input-diff payloads, and generation prints a uv add command and the next steps to stderr. generate() takes them as generate_server and the server_* options, and returns every generated file without an output, writing no stubs. It needs Python 3.11 or later, both to run and as --target-python-version, --use-missing-sentinel for pydantic_v2.dataclass models, --target-pydantic-version 2.14 with --use-missing-sentinel, and --openapi-scopes with api. A setting left unset, by the options, pyproject.toml, generate(), and the preset, takes that value, and the command line prints how to set it; one set otherwise is refused.

cli-option.install-skill¶

• Kind: cli-option
• Target: --install-skill, --skill-scope, and --overwrite-skill
• Since: 0.76.0

Bundled Agent Skill installation is experimental; supported clients, scopes, and installed skill contents may change.

Use --install-skill with codex or claude-code. --skill-scope selects a project or user installation, and --overwrite-skill safely replaces an existing regular skill directory.

cli-option.schema-validator-type¶

• Kind: cli-option
• Target: --schema-validator-type
• Since: 0.66.1

Schema-derived runtime validator backend selection is experimental and may change as validation backends are added.

The only currently implemented backend is 'pydantic-v2', which preserves the existing generated Pydantic v2 validator behavior.

cli-option.use-missing-sentinel¶

• Kind: cli-option
• Target: --use-missing-sentinel
• Since: 0.66.1

Pydantic MISSING sentinel output is experimental because it depends on pydantic.experimental.missing_sentinel, or on pydantic.MISSING for a 2.14 target.

The option requires Pydantic v2 BaseModel or dataclass output and a target Pydantic version that supports the MISSING sentinel. Server and client generation require it for Pydantic v2 dataclass output, and --target-pydantic-version 2.14 with it.

cli-option.use-type-alias¶

• Kind: cli-option
• Target: --use-type-alias
• Since: 0.36.0

Type alias output is experimental and may change as Python typing support evolves.

The option replaces root model classes with type aliases where possible. Pydantic v2 output may use TypeAliasType or Python 3.12 type statements depending on the target Python version.

cli-option.use-type-alias-type¶

• Kind: cli-option
• Target: --use-type-alias-type
• Since: 0.71.1

Runtime TypeAliasType output is experimental and may change as Python typing support evolves.

The option implies --use-type-alias and selects TypeAliasType for Python 3.10 and 3.11. Python 3.12 and newer continue to use native type statements.

extra.httpx2¶

• Kind: extra
• Target: datamodel-code-generator[httpx2]
• Since: 0.71.1

The HTTPX2-backed HTTP client is experimental and may change as compatibility is validated.

datamodel-code-generator[http] remains the stable HTTPX backend and is not deprecated; datamodel-code-generator[httpx2] is experimental. The default HTTP backend policy is auto: stable httpx is selected when its client module is installed, including when both pairs are installed, and experimental httpx2 is selected only when that module is absent. Use --http-backend httpx2 or HTTPBackend.HTTPX2 to require the experimental pair. Explicit selections and paired dependency errors do not fall back.

formatter.builtin¶

• Kind: formatter
• Target: --formatters builtin
• Since: 0.59.0

The internal formatter is experimental and may change as generated-output coverage is expanded.

The formatter is designed for generated model modules and is not a general-purpose Python formatter.

input-format.asyncapi¶

• Kind: input-format
• Target: --input-file-type asyncapi
• Since: 0.59.0

AsyncAPI input support is experimental and may change as real-world usage is validated.

The parser focuses on message payload model generation from AsyncAPI documents.

input-format.avro¶

• Kind: input-format
• Target: --input-file-type avro
• Since: 0.59.0

Apache Avro schema input support is experimental and may change as real-world usage is validated.

The parser generates Python models from Avro schemas; it does not provide Avro runtime validation.

input-format.mcp-tools¶

• Kind: input-format
• Target: --input-file-type mcp-tools
• Since: 0.60.0

MCP tool schema profile input support is experimental and may change as MCP schemas evolve.

The input is converted from MCP tool inputSchema/outputSchema entries into JSON Schema definitions before model generation.

input-format.protobuf¶

• Kind: input-format
• Target: --input-file-type protobuf
• Since: 0.59.0

Protocol Buffers input support is experimental and may change as real-world usage is validated.

The parser generates Python models from .proto schemas; it does not provide protobuf runtime validation or gRPC code generation.

input-format.xmlschema¶

• Kind: input-format
• Target: --input-file-type xmlschema
• Since: 0.59.0

XML Schema input support is experimental and may change as real-world usage is validated.

The parser focuses on model generation from XSD documents, not full XML instance validation.