Output Plugins

Output type plugins define structured schemas for task output validation. Tasks reference these types to ensure the assistant produces correctly formatted content.

Directory Layout

my_outputs/
    manifest.py
    outputs.py

Example

manifest.py:

from app.plugins.base import PluginManifest, PluginType

manifest = PluginManifest(
    name="my_outputs",
    plugin_type=PluginType.OUTPUT,
    version="1.0.0",
    description="Custom output type definitions",
    author="Your Name",
)

outputs.py:

from pydantic import BaseModel

from app.plugins.base import OutputType


class NewsletterSchema(BaseModel):
    """Structured output for email newsletters."""

    title: str
    subject_line: str
    body: str
    preview_text: str


class SocialPostSchema(BaseModel):
    """Structured output for social media posts."""

    platform: str
    text: str
    hashtags: list[str] = []


CORE_OUTPUT_TYPES: list[OutputType] = [
    OutputType(
        name="newsletter",
        label="Newsletter",
        description="An email newsletter with subject line and preview text.",
        schema=NewsletterSchema,
    ),
    OutputType(
        name="social_post",
        label="Social Post",
        description="A social media post with platform and hashtags.",
        schema=SocialPostSchema,
    ),
]

How It Works

The output types list must be named CORE_OUTPUT_TYPES
Each OutputType has a Pydantic BaseModel schema that TeamWeb AI uses to validate the assistant’s structured output
Output names must be unique across all plugins
Output plugins do not require a config_schema since there is nothing to configure

`OutputType` Fields

Field	Description
`name`	Unique identifier (e.g. `"newsletter"`)
`label`	Human-readable display name (e.g. “Newsletter”)
`description`	Short description shown when selecting an output type
`schema`	Pydantic `BaseModel` class for validating the output

Built-in Output Types

TeamWeb AI ships with these output types in the core_outputs plugin: blog_post, file, research, summary, proofreading, image, decision, and freeform. Custom output plugins can add to this list but should avoid reusing these names.

Provider Plugins