Skip to main content

​
Overview

Recipe endpoints provide insights into model lineage and dependencies. They help you understand how bakes relate to each other, what resources are needed to recreate a bake, and visualize the complete model evolution tree.
SDK Guide: For detailed SDK examples, use cases, and workflows, see the Model Lineage & Recipes guide in the SDK section.

​
Endpoints

​
Get Dependency Graph

Get the dependency graph for a specific bake. Returns all resources (bakes, targets, prompts) and their dependencies needed to recreate this bake, including parent bakes and all transitive dependencies. Endpoint: GET /v1/repo/{repo_name}/recipe/{bake_name}/dependency-graph Request: 
curl -X GET "https://bapi.bread.com.ai/v1/repo/my_repo/recipe/my_bake/dependency-graph" \
  -H "Authorization: Bearer $BREAD_API_KEY"
SDK Usage:
  • TypeScript: client.recipes.getDependencyGraph('my_bake', { repo_name: 'my_repo' })
  • Python: client.recipes.get_dependency_graph('my_bake', repo_name='my_repo')
See SDK Guide for detailed examples.
​
repo_name
string
required
Repository name
​
bake_name
string
required
Bake name
Response: 200 OK 
{
  "base_model": "Qwen/Qwen3-32B",
  "bakes": {
    "parent_bake": {
      "datasets": [{"target": "coding_target", "weight": 1.0}],
      "epochs": 3
    },
    "my_bake": {
      "datasets": [{"target": "advanced_target", "weight": 1.0}],
      "checkpoint": "parent_bake"
    }
  },
  "targets": {
    "coding_target": {
      "model_name": "Qwen/Qwen3-32B",
      "teacher_prompt": "system_prompt",
      "student_prompt": "user_prompt"
    },
    "advanced_target": {
      "model_name": "Qwen/Qwen3-32B",
      "teacher_prompt": "system_prompt_advanced",
      "student_prompt": "user_prompt_advanced"
    }
  },
  "prompts": {
    "system_prompt": "prompts/system_prompt.yml",
    "user_prompt": "prompts/user_prompt.yml",
    "system_prompt_advanced": "prompts/system_prompt_advanced.yml",
    "user_prompt_advanced": "prompts/user_prompt_advanced.yml"
  },
  "edges": [
    ["bake", "my_bake", "bake", "parent_bake"],
    ["bake", "parent_bake", "target", "coding_target"],
    ["bake", "my_bake", "target", "advanced_target"],
    ["target", "coding_target", "prompt", "system_prompt"],
    ["target", "coding_target", "prompt", "user_prompt"]
  ]
}
Response Fields:
  • base_model: string - Base model name
  • bakes: object - Dictionary of bake configs
  • targets: object - Dictionary of target configs
  • prompts: object - Dictionary of prompt names to file paths
  • edges: Array<[source_type, source_name, target_type, target_name]> - List of dependency edges
The graph is built using BFS traversal starting from the specified bake.

​
Get Recreation Plan

Get a step-by-step plan to recreate a bake, including all dependencies. The plan includes all prompts, targets, and bakes needed, in the correct execution order. Endpoint: GET /v1/repo/{repo_name}/recipe/{bake_name}/recreation-plan Request: 
curl -X GET "https://bapi.bread.com.ai/v1/repo/my_repo/recipe/my_bake/recreation-plan" \
  -H "Authorization: Bearer $BREAD_API_KEY"
SDK Usage:
  • TypeScript: client.recipes.getRecreationPlan('my_bake', { repo_name: 'my_repo' })
  • Python: client.recipes.get_recreation_plan('my_bake', repo_name='my_repo')
See SDK Guide for detailed examples.
​
repo_name
string
required
Repository name
​
bake_name
string
required
Bake name
Response: 200 OK 
{
  "base_model": "Qwen/Qwen3-32B",
  "total_steps": 5,
  "steps": [
    {
      "step": 1,
      "type": "prompt",
      "name": "system_prompt",
      "config": {
        "messages": [
          {"role": "system", "content": "You are a helpful assistant."}
        ]
      }
    },
    {
      "step": 2,
      "type": "prompt",
      "name": "user_prompt",
      "config": {
        "messages": [
          {"role": "user", "content": "Complete this task: {task}"}
        ]
      }
    },
    {
      "step": 3,
      "type": "target",
      "name": "coding_target",
      "config": {
        "model_name": "Qwen/Qwen3-32B",
        "teacher_prompt": "system_prompt",
        "student_prompt": "user_prompt"
      }
    },
    {
      "step": 4,
      "type": "bake",
      "name": "parent_bake",
      "config": {
        "datasets": [{"target": "coding_target", "weight": 1.0}],
        "epochs": 3
      }
    },
    {
      "step": 5,
      "type": "bake",
      "name": "my_bake",
      "config": {
        "datasets": [{"target": "advanced_target", "weight": 1.0}],
        "checkpoint": "parent_bake"
      }
    }
  ],
  "resources": {
    "prompts": ["system_prompt", "user_prompt"],
    "targets": ["coding_target", "advanced_target"],
    "bakes": [
      {
        "name": "parent_bake",
        "model_name": "user/repo/parent_bake/42"
      },
      {
        "name": "my_bake",
        "model_name": "user/repo/my_bake/35"
      }
    ]
  }
}
Response Fields:
  • base_model: string - Base model name (e.g., ‘Qwen/Qwen3-32B’)
  • total_steps: number - Total number of steps
  • steps: Array<RecreationPlanStep> - Ordered list of steps to recreate the bake
    • step: number - Step number
    • type: string - Resource type: ‘prompt’, ‘target’, or ‘bake’
    • name: string - Resource name
    • config: object - Resource configuration
  • resources: object - Summary of all resources
    • prompts: Array<string> - List of prompt names
    • targets: Array<string> - List of target names
    • bakes: Array<BakeResource> - List of bake resources with model names
The plan is topologically sorted to ensure dependencies are created before dependents. Parent bake checkpoints are assumed to use the final checkpoint from each bake. All configurations are cleaned and ready for use (no internal file paths).

​
Get Repo Tree

Get the complete model lineage tree for a repository. Returns all bakes in the repository with their parent-child relationships, status, checkpoints, and full model paths. Endpoint: GET /v1/repo/{repo_name}/tree Request: 
curl -X GET "https://bapi.bread.com.ai/v1/repo/my_repo/tree" \
  -H "Authorization: Bearer $BREAD_API_KEY"
SDK Usage:
  • TypeScript: client.repo.getTree('my_repo')
  • Python: client.repo.get_tree('my_repo')
See SDK Guide for detailed examples.
​
repo_name
string
required
Repository name
Response: 200 OK 
{
  "base_model": "Qwen/Qwen3-32B",
  "bakes": {
    "parent_bake": {
      "status": "complete",
      "model_name": ["user/repo/parent_bake/42"],
      "checkpoints": [42],
      "config": {
        "datasets": [{"target": "coding_target", "weight": 1.0}],
        "epochs": 3
      }
    },
    "my_bake": {
      "status": "running",
      "model_name": [],
      "checkpoints": [],
      "config": {
        "datasets": [{"target": "advanced_target", "weight": 1.0}],
        "epochs": 5
      }
    }
  },
  "edges": [
    ["bake", "my_bake", "bake", "parent_bake"]
  ]
}
Response Fields:
  • base_model: string - Base model name (e.g., ‘Qwen/Qwen3-32B’)
  • bakes: object - Dictionary of all bakes in the repository, keyed by bake name
    • Each bake has:
      • status: string - Bake status: ‘complete’, ‘failed’, ‘running’, ‘pending’, or ‘unknown’
      • model_name: Array<string> - List of full model paths with checkpoints (e.g., ["user/repo/bake_name/checkpoint"])
      • checkpoints: Array<number> - List of checkpoint numbers
      • config: object - Bake configuration (bake.yml)
  • edges: Array<[source_type, source_name, target_type, target_name]> - List of parent-child edges representing model lineage relationships
This provides a complete view of model evolution without requiring a specific starting bake. Unlike dependency-graph which starts from a specific bake, this endpoint returns the entire repository’s model lineage tree in a single call.