Conscribe

Inheritance is registration. __init__ signature is config schema.

Conscribe is a Python library that provides automatic class registration and config typing stub generation for layered architectures. It eliminates two categories of boilerplate:

1. Manual registration — Write a class, inherit a base → it's registered. No registry["foo"] = FooClass.
2. Config guesswork — Your __init__ parameters become the config schema. IDE autocomplete and fail-fast validation come for free.

pip install conscribe

Requires Python >= 3.9. Built on Pydantic v2.

Who Is This For?

Framework developers building config-driven systems with pluggable layers (agents, LLM providers, browser backends, etc.) who need registries, factories, protocol checks, and config schemas — without N layers × boilerplate.

Framework users who write YAML configs and want IDE autocomplete, parameter docs, and fail-fast validation at startup instead of crashing N steps in.

Quick Start

1. Define a Layer

from typing import Protocol, runtime_checkable
from conscribe import create_registrar

@runtime_checkable
class ChatModelProtocol(Protocol):
    def chat(self, messages: list[dict]) -> str: ...

LLMRegistrar = create_registrar(
    "llm",
    ChatModelProtocol,
    discriminator_field="provider",
    strip_prefixes=["Chat"],
)

2. Create a Base Class

class ChatBaseModel(metaclass=LLMRegistrar.Meta):
    __abstract__ = True

3. Write Implementations (auto-registered)

class ChatOpenAI(ChatBaseModel):
    """OpenAI LLM provider.

    Args:
        model_id: Model identifier, e.g. gpt-4o
        temperature: Sampling temperature, 0-2
    """
    def __init__(self, *, model_id: str, temperature: float = 0.0):
        self.model_id = model_id
        self.temperature = temperature

    def chat(self, messages): ...

# Registered as "open_ai". No decorator. No registry call.

4. Discover & Use

from conscribe import discover

discover("my_app.llm.providers")

llm_cls = LLMRegistrar.get("open_ai")   # → ChatOpenAI
llm = llm_cls(model_id="gpt-4o")
print(LLMRegistrar.keys())              # ["open_ai", "anthropic", ...]

Config Typing

Your __init__ signature is the config schema. Conscribe extracts it, builds a Pydantic discriminated union, and generates stubs for IDE autocomplete:

from conscribe import build_layer_config, generate_layer_config_source

result = build_layer_config(LLMRegistrar)
source = generate_layer_config_source(result)

See the Config Typing Guide for full details.

Config Tiers

Nested Config (Hierarchical Keys)

For layers with natural hierarchies (e.g., model_type → provider):

LLM = create_registrar(
    "llm", ChatModelProtocol,
    discriminator_fields=["model_type", "provider"],
    key_separator=".",
)

Produces hybrid YAML configs where level 0 is flat and level 1+ is nested:

llm:
  model_type: openai        # flat (level 0)
  temperature: 0.7          # flat (level 0 param)
  provider:                 # nested (level 1)
    name: azure
    deployment: my-deploy

Cross-Registry Wiring

Declare which values from other registries a class accepts. Conscribe constrains the config field to Literal[...]:

# Given: LoopRegistrar with "react" and "codeact" registered

class BaseAgent(metaclass=AgentRegistrar.Meta):
    __abstract__ = True
    __wiring__ = {"loop": "loop"}  # all keys from "loop" registry

class SWEAgent(BaseAgent):
    __wiring__ = {"loop": ("loop", ["react"])}  # narrows to subset

    def __init__(self, *, max_steps: int = 10): ...

Generated config:

class SweAgentConfig(BaseModel):
    name: Literal["swe_agent"] = "swe_agent"
    max_steps: int = 10
    loop: Literal["react"] = ...  # wired from: loop

Four modes:

• "loop": "agent_loop" — auto-discover all keys from registry
• "loop": ("agent_loop", ["react", "codeact"]) — explicit subset
• "obs": ("observation", ["terminal"], ["filesystem"]) — required + optional keys
• "browser": ["chromium", "firefox"] — literal list (no registry)

The 3-element tuple distinguishes required and optional keys — both appear in Literal[...], but the distinction is available as metadata for downstream negotiation logic.

Inheritance: child __wiring__ merges with parent (child keys override). Use None to exclude: {"llm": None}.

Composed Config (Multi-Layer Inline Wiring)

Combine multiple layers into a single config schema. Wired fields become full inline config objects instead of key selectors — enabling recursive IDE autocompletion across layers:

from conscribe import build_composed_config, generate_composed_json_schema

result = build_composed_config(
    {"llm": LLMRegistrar, "agent": AgentRegistrar},
    inline_wiring=True,   # wired fields become target layer union types
)
schema = generate_composed_json_schema(result)

YAML config with inline wiring:

agents:
  - name: browser_use
    use_vision: true
    llm:                        # full LLM config inline — IDE autocompletes all fields
      provider: openai
      model: gpt-4o
      temperature: 0.7
  - name: react
    llm:
      provider: anthropic
      model: claude-3

Use inline_wiring=False to keep wired fields as Literal[...] key selectors.

Cross-Registry Diamond Inheritance

Register a class in multiple registries:

CombinedMeta = LLM.Meta | Agent.Meta

class LLMAgent(metaclass=CombinedMeta):
    ...  # in both LLM and Agent registries

API Reference

Registration

Config Typing

Design Principles

• Zero registration burden — Inherit a base class = registered
• __init__ is the single source of truth — No duplicate config definitions
• Fail-fast — Duplicate keys raise immediately; invalid config rejects at startup
• Domain-agnostic — Pure infrastructure, knows nothing about agents or LLMs
• Stubs and runtime are separate — Stale stubs don't affect correctness

Documentation

Full documentation is shipped inside the package (accessible at site-packages/conscribe/) and browsable on GitHub:

License

MIT