refactor(config): make a single source of truth file (#524)

* refactor

* fix nits

* add get from env

* refactor logic

agenthub/monologue_agent/utils/memory.py

@@ -7,7 +7,7 @@ from llama_index.vector_stores.chroma import ChromaVectorStore
 from opendevin import config
 from . import json
 
-embedding_strategy = config.get_or_default("LLM_EMBEDDING_MODEL", "local")
+embedding_strategy = config.get("LLM_EMBEDDING_MODEL")
 
 # TODO: More embeddings: https://docs.llamaindex.ai/en/stable/examples/embeddings/OpenAI/
 # There's probably a more programmatic way to do this.
@@ -15,7 +15,7 @@ if embedding_strategy == "llama2":
     from llama_index.embeddings.ollama import OllamaEmbedding
     embed_model = OllamaEmbedding(
         model_name="llama2",
-        base_url=config.get_or_default("LLM_BASE_URL", "http://localhost:8000"),
+        base_url=config.get_or_error("LLM_BASE_URL"),
         ollama_additional_kwargs={"mirostat": 0},
     )
 elif embedding_strategy == "openai":

opendevin/config.py

@@ -5,12 +5,33 @@ from dotenv import load_dotenv
 
 load_dotenv()
 
+DEFAULT_CONFIG = {
+    "LLM_API_KEY": None,
+    "LLM_BASE_URL": None,
+    "WORKSPACE_DIR": os.path.join(os.getcwd(), "workspace"),
+    "LLM_MODEL": "gpt-4-0125-preview",
+    "SANDBOX_CONTAINER_IMAGE": "ghcr.io/opendevin/sandbox",
+    "RUN_AS_DEVIN": "false",
+    "LLM_EMBEDDING_MODEL": "local",
+    "LLM_NUM_RETRIES": 6,
+    "LLM_COOLDOWN_TIME" : 1,
+    "DIRECTORY_REWRITE" : "",
+    "PROMPT_DEBUG_DIR": "",
+}
+
 config_str = ""
 if os.path.exists("config.toml"):
     with open("config.toml", "rb") as f:
         config_str = f.read().decode("utf-8")
 
-config = toml.loads(config_str)
+tomlConfig = toml.loads(config_str)
+config = DEFAULT_CONFIG.copy()
+for key, value in config.items():
+  if key in os.environ:
+    config[key] = os.environ[key]
+  elif key in tomlConfig:
+    config[key] = tomlConfig[key]
+
 
 def _get(key: str, default):
     value = config.get(key, default)
@@ -38,3 +59,9 @@ def get_or_none(key: str):
     Get a key from the config, or return None if it doesn't exist.
     """
     return _get(key, None)
+
+def get(key: str):
+    """
+    Get a key from the config, please make sure it exists.
+    """
+    return config.get(key)

opendevin/llm/llm.py

@@ -6,12 +6,12 @@ from functools import partial
 
 from opendevin import config
 
-DEFAULT_MODEL_NAME = config.get_or_default("LLM_MODEL", "gpt-4-0125-preview")
-DEFAULT_API_KEY = config.get_or_none("LLM_API_KEY")
-DEFAULT_BASE_URL = config.get_or_none("LLM_BASE_URL")
-DEFAULT_LLM_NUM_RETRIES = config.get_or_default("LLM_NUM_RETRIES", 6)
-DEFAULT_LLM_COOLDOWN_TIME = config.get_or_default("LLM_COOLDOWN_TIME", 1)
-PROMPT_DEBUG_DIR = config.get_or_default("PROMPT_DEBUG_DIR", "")
+DEFAULT_API_KEY = config.get("LLM_API_KEY")
+DEFAULT_BASE_URL = config.get("LLM_BASE_URL")
+DEFAULT_MODEL_NAME = config.get("LLM_MODEL")
+DEFAULT_LLM_NUM_RETRIES = config.get("LLM_NUM_RETRIES")
+DEFAULT_LLM_COOLDOWN_TIME = config.get("LLM_COOLDOWN_TIME")
+PROMPT_DEBUG_DIR = config.get("PROMPT_DEBUG_DIR")
 
 class LLM:
     def __init__(self,

opendevin/main.py

@@ -25,7 +25,7 @@ def parse_arguments():
     parser.add_argument("-t", "--task", type=str, default="", help="The task for the agent to perform")
     parser.add_argument("-f", "--file", type=str, help="Path to a file containing the task. Overrides -t if both are provided.")
     parser.add_argument("-c", "--agent-cls", default="MonologueAgent", type=str, help="The agent class to use")
-    parser.add_argument("-m", "--model-name", default=config.get_or_default("LLM_MODEL", "gpt-4-0125-preview"), type=str, help="The (litellm) model name to use")
+    parser.add_argument("-m", "--model-name", default=config.get("LLM_MODEL"), type=str, help="The (litellm) model name to use")
     parser.add_argument("-i", "--max-iterations", default=100, type=int, help="The maximum number of iterations to run the agent")
     return parser.parse_args()
 

opendevin/sandbox/sandbox.py

@@ -15,15 +15,12 @@ from opendevin import config
 InputType = namedtuple("InputType", ["content"])
 OutputType = namedtuple("OutputType", ["content"])
 
-DIRECTORY_REWRITE = config.get_or_default(
-    "DIRECTORY_REWRITE", ""
-)  # helpful for docker-in-docker scenarios
-
-CONTAINER_IMAGE = config.get_or_default("SANDBOX_CONTAINER_IMAGE", "ghcr.io/opendevin/sandbox")
+DIRECTORY_REWRITE = config.get("DIRECTORY_REWRITE")  # helpful for docker-in-docker scenarios
+CONTAINER_IMAGE = config.get("SANDBOX_CONTAINER_IMAGE")
 
 # FIXME: On some containers, the devin user doesn't have enough permission, e.g. to install packages
 # How do we make this more flexible?
-RUN_AS_DEVIN = config.get_or_default("RUN_AS_DEVIN", "false").lower() != "false"
+RUN_AS_DEVIN = config.get("RUN_AS_DEVIN").lower() != "false"
 USER_ID = 1000
 if config.get_or_none("SANDBOX_USER_ID") is not None:
     USER_ID = int(config.get_or_default("SANDBOX_USER_ID", ""))

opendevin/server/session.py

@@ -15,11 +15,11 @@ from opendevin.controller import AgentController
 from opendevin.llm.llm import LLM
 from opendevin.observation import Observation, UserMessageObservation
 
-DEFAULT_API_KEY = config.get_or_none("LLM_API_KEY")
-DEFAULT_BASE_URL = config.get_or_none("LLM_BASE_URL")
-DEFAULT_WORKSPACE_DIR = config.get_or_default("WORKSPACE_DIR", os.path.join(os.getcwd(), "workspace"))
-LLM_MODEL = config.get_or_default("LLM_MODEL", "gpt-4-0125-preview")
-CONTAINER_IMAGE = config.get_or_default("SANDBOX_CONTAINER_IMAGE", "ghcr.io/opendevin/sandbox")
+DEFAULT_API_KEY = config.get("LLM_API_KEY")
+DEFAULT_BASE_URL = config.get("LLM_BASE_URL")
+DEFAULT_WORKSPACE_DIR = config.get("WORKSPACE_DIR")
+LLM_MODEL = config.get("LLM_MODEL")
+CONTAINER_IMAGE = config.get("SANDBOX_CONTAINER_IMAGE")
 
 class Session:
     """Represents a session with an agent.