Ananke.Design 0.1.0

Ananke.Design

Design-time tooling for Ananke — YAML manifest import, workflow topology DSL, model resolution, and Mermaid diagram export. Define pipeline structure declaratively; bind job implementations in code.

Install

dotnet add package Ananke.Design

What it does

Ananke.Design separates what the workflow looks like (graph structure, model aliases, system prompts) from what each job does (C# code). This enables design-first workflows, LLM-generated topologies, and external configuration — while keeping job implementations type-safe and testable.

Quick start — DSL scaffold

Parse a topology from text, bind implementations, build a runnable workflow:

using Ananke.Design;

var scaffold = WorkflowScaffold.Parse<MyState>("etl-pipeline", """
    plan -> fork(fetch_a, fetch_b)
    fetch_a -> transform_a
    fetch_b -> transform_b
    join(transform_a, transform_b) -> combine
    combine -> End
    """);

var workflow = scaffold
    .Bind("plan", async (state, ct) => state with { Step = "planned" })
    .Bind("fetch_a", fetchAJob)
    .Bind("fetch_b", fetchBJob)
    .Bind("transform_a", async (state, ct) => state)
    .Bind("transform_b", async (state, ct) => state)
    .Bind("combine", async (state, ct) => state)
    .BindMerge("combine", branches => branches[0])
    .Build();

var result = await workflow.RunAsync(initialState);

Quick start — YAML manifest

Load a full manifest with model definitions, agent jobs, and connections from an .ananke.yml file:

name: research-pipeline

models:
  fast:
    provider: openai
    model: gpt-4.1-mini
  local:
    provider: openai
    model: llama3.2
    endpoint: http://localhost:11434/v1

jobs:
  plan:
    type: agent
    model: fast
    system_prompt: You are a research planner.
  search:
    type: agent
    model: local
  synthesize:
    type: code

connections:
  - plan -> fork(search_web, search_db)
  - join(search_web, search_db) -> synthesize
  - synthesize -> End

var manifest = WorkflowManifest.Load("pipeline.ananke.yml");

var models = new ModelResolver()
    .Register("openai", "OpenAI", OpenAIChatAgentModel.Create)
    .Resolve(manifest, key => config[key]);

DSL syntax

Full syntax reference: docs/workflow-dsl.md

What is included

Mermaid diagram export

var definition = workflow.Validate();
var mermaid = definition.ToMermaid();
Console.WriteLine(mermaid);

Outputs:

graph TD
    _plan["▶ plan"]
    _fetch_a["fetch_a"]
    _fetch_b["fetch_b"]
    _combine["combine"]
    _end(["End"])

    _plan -->|fork| _fetch_a
    _plan -->|fork| _fetch_b
    _fetch_a --> _combine
    _fetch_b --> _combine
    _combine --> _end

Related packages

Documentation

Full docs, demos, and architecture: github.com/sevensamurai/Ananke

License

Apache 2.0