deepgoal 0.0.3

A pre-alpha AI Agent framework for goal-driven automatic programming.

中文版本

DeepGoal

DeepGoal is a pre-alpha AI Agent framework for goal-driven automatic programming. It models engineering workflows as composable pipelines: container nodes orchestrate the flow, while executors perform the leaf work.

Status

DeepGoal is under active design and development. APIs may change before the first stable release.

Requirements

• Python 3.11+

Installation

pip install deepgoal

For local development:

pip install -e ".[dev]"

Minimal Example

from deepgoal.core.executor import Executor
from deepgoal.core.node import PipelineInput, PipelineOutput, PipelineResult
from deepgoal.core.pipeline.pipe import Pipe
from deepgoal.core.types import EngineOptions


class WriteSpecExecutor(Executor):
    async def execute(self, input: PipelineInput, options: EngineOptions) -> PipelineResult:
        return PipelineResult.ok(PipelineOutput(primary_path="workspace/spec.md"))


pipe = Pipe(items=[WriteSpecExecutor()])
result = await pipe.run(PipelineInput(primary_path="goal.md"))
print(result.output.primary_path)

Core Idea

• Node / Pipe are containers that expose pipeline input and output boundaries.
• Executor is a leaf execution unit.
• Containers can compose child nodes and executors into serial, parallel, loop, or branch workflows.