模型编排

Installation

pip install model-compose

Or install from source:

git clone https://github.com/hanyeol/model-compose.git
cd model-compose
pip install -e .

Requires: Python 3.10 or higher

---

Deploy in Any Runtime

Run in native, process, Docker, or native container mode. The same configuration works across all runtimes — switch with one line.

controller:
  runtime:
    type: docker
    image: my-ai-service:latest
    ports:
      - "8080:8080"
  adapter:
    type: http-server
    port: 8080

Setup for development

git clone https://github.com/hanyeol/model-compose.git cd model-compose pip install -e . ```

---

Quick Start

Define your AI runtime in a model-compose.yml:

controller:
  adapter:
    type: http-server
    port: 8080
  webui:
    port: 8081

workflows:
  - id: chat
    default: true
    jobs:
      - component: chatgpt

components:
  - id: chatgpt
    type: http-client
    base_url: https://api.openai.com/v1
    path: /chat/completions
    method: POST
    headers:
      Authorization: Bearer ${env.OPENAI_API_KEY}
    body:
      model: gpt-4o
      messages:
        - role: user
          content: ${input.prompt}

Create a .env file:

OPENAI_API_KEY=your-key

Run it:

model-compose up

Your AI runtime is now serving at http://localhost:8080 with Web UI at http://localhost:8081.

Explore examples for more workflows or read the Documentation.

---

Declarative YAML Configuration

Define your entire AI system in a single YAML file. Workflows, agents, models, APIs, vector/graph stores, and runtimes — all composed and deployed together without custom code.

controller:
  adapter:
    type: http-server
    port: 8080

workflows:
  - id: chat
    default: true
    jobs:
      - component: chatgpt

components:
  - id: chatgpt
    type: http-client
    base_url: https://api.openai.com/v1
    action:
      path: /chat/completions
      method: POST

Flexible Component System

20+ reusable component types. Mix HTTP clients, local models, vector stores, shell commands, and workflows in any combination. Define once, use everywhere.

components:
  - id: chatgpt
    type: http-client

  - id: local-llm
    type: model

  - id: assistant
    type: agent

  - id: knowledge
    type: vector-store

  - id: cache
    type: key-value-store

  - id: runner
    type: shell

Advanced Workflow Composition

Chain jobs with conditional logic, parallel execution, and data transformation. Pass data between jobs with variable binding — ${input}, ${response}, ${env} — with type conversion and defaults.

workflows:
  - id: rag-pipeline
    jobs:
      - id: embed
        component: embedder
        input:
          text: ${input.query}

      - id: search
        component: vector-store
        action: search
        input:
          vector: ${jobs.embed.output}
        depends_on: [embed]

      - id: answer
        component: chatgpt
        input:
          context: ${jobs.search.output}
          question: ${input.query}
        depends_on: [search]

AI Agent Components

Build autonomous AI agents that use workflows as tools. Agents reason, plan, and execute multi-step tasks by dynamically invoking other workflows — all defined declaratively in YAML.

components:
  - id: research-agent
    type: agent
    tools:
      - search-web
      - fetch-page
    max_iteration_count: 10
    action:
      model:
        component: chatgpt
        input:
          messages: ${messages}
          tools: ${tools}
      system_prompt: You are a web research assistant.
      user_prompt: ${input.question}

Universal AI Service Integration

Connect to OpenAI, Anthropic, Google, xAI, ElevenLabs, and any custom HTTP API. Mix and match providers in a single workflow.

components:
  - id: claude
    type: http-client
    base_url: https://api.anthropic.com/v1
    action:
      path: /messages
      method: POST
      headers:
        x-api-key: ${env.ANTHROPIC_API_KEY}
        anthropic-version: "2023-06-01"
      body:
        model: claude-opus-4-20250514
        max_tokens: 1024
        messages:
          - role: user
            content: ${input.prompt}

Built-in Data Store Integration

Native integration with Chroma, FAISS, Milvus, Qdrant for vector search. Neo4j and ArangoDB for graph stores. Redis for key-value storage. Build RAG systems with embedding search and semantic retrieval.

components:
  - id: knowledge
    type: vector-store
    driver: chroma
    actions:
      - id: insert
        collection: docs
        method: insert
        vector: ${input.vector}
        metadata:
          text: ${input.text}

      - id: search
        collection: docs
        method: search
        query: ${input.vector}

Distributed Workflow Execution

Scale AI workloads across multiple machines using Redis-backed queue dispatch. Add workers to scale horizontally without shared filesystem or code changes.

controller:
  adapter:
    type: http-server
    port: 8080
  queue:
    driver: redis
    host: localhost
    port: 6379
    name: my-queue