reorganize and add new services for propmt

docker-compose.yml

@@ -214,15 +214,25 @@ services:
             file: ./services/completion-stateless-sdk/docker-compose.yml
             service: completion-stateless-sdk
 
-    completion-live-sdk:
+    completion-new-sdk:
         extends:
-            file: ./services/completion-live-sdk/docker-compose.yml
-            service: completion-live-sdk
+            file: ./services/completion-new-sdk/docker-compose.yml
+            service: completion-new-sdk
 
-    chat-live-sdk:
+    chat-new-sdk:
         extends:
-            file: ./services/chat-live-sdk/docker-compose.yml
-            service: chat-live-sdk
+            file: ./services/chat-new-sdk/docker-compose.yml
+            service: chat-new-sdk
+
+    completion-new-sdk-prompt:
+        extends:
+            file: ./services/completion-new-sdk-prompt/docker-compose.yml
+            service: completion-new-sdk-prompt
+
+    chat-new-sdk-prompt:
+        extends:
+            file: ./services/chat-new-sdk-prompt/docker-compose.yml
+            service: chat-new-sdk-prompt
 
 networks:
     agenta-network:

services/README.md

@@ -0,0 +1,33 @@
+# Agenta Services
+
+This directory contains various versions of Agenta's LLM services, each offering distinct capabilities and interfaces for language model interactions.
+
+## Service Overview
+
+### Legacy Services
+- **completion-old-sdk**: Original completion service (as in current release)
+- **chat-old-sdk**: Original chat service (as in current release)
+
+### New Services
+All services with "new-sdk" utilize the modified SDK, which includes these changes:
+- Configuration is now nested under `agenta_config` in the request body (no longer flattened)
+- Implements the stateless SDK (no interface changes, but may introduce future issues in cloud deployment due to lack of testing)
+
+We've created two versions of each new service:
+1. Original logic with new SDK:
+   - completion-new-sdk
+   - chat-new-sdk
+2. New prompt object and updated logic:
+   - completion-new-sdk-prompt
+   - chat-new-sdk-prompt
+
+## Service Components
+
+Each service includes:
+- Docker configuration (`docker-compose.yml`)
+- REST API documentation (`.rest` files)
+- Implementation code (`_app.py`)
+
+## Usage
+
+For usage examples and API details, refer to the `.rest` files in each service's directory.

services/chat-new-sdk-prompt.rest

@@ -0,0 +1,33 @@
+### Test chat-new-sdk-prompt
+POST http://localhost/chat-new-sdk-prompt/chat
+Content-Type: application/json
+
+{
+    "inputs": {
+        "message": "What is the capital of France?"
+    }
+}
+
+### Test chat configuration with prompt
+POST http://localhost/chat-new-sdk-prompt/configure
+Content-Type: application/json
+
+{
+    "model": "gpt-3.5-turbo",
+    "temperature": 0.7,
+    "max_tokens": 100,
+    "prompt": {
+        "messages": [
+            {
+                "role": "system",
+                "content": "You are a helpful assistant specializing in geography."
+            }
+        ],
+        "template_format": "fstring",
+        "response_format": {
+            "type": "text"
+        },
+        "tools": null,
+        "tool_choice": "auto"
+    }
+}

services/chat-new-sdk-prompt/docker-compose.yml

@@ -0,0 +1,23 @@
+services:
+  chat-new-sdk-prompt:
+    build: .
+    volumes:
+      - .:/app
+      - ../../agenta-cli:/agenta-cli
+    environment:
+      - AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED=True
+      - AGENTA_HOST=http://host.docker.internal
+    networks:
+      - agenta-network
+    labels:
+      - "traefik.http.routers.chat-new-sdk-prompt.rule=PathPrefix(`/chat-new-sdk-prompt/`)"
+      - "traefik.http.routers.chat-new-sdk-prompt.entrypoints=web"
+      - "traefik.http.middlewares.chat-new-sdk-prompt-strip.stripprefix.prefixes=/chat-new-sdk-prompt"
+      - "traefik.http.middlewares.chat-new-sdk-prompt-strip.stripprefix.forceslash=true"
+      - "traefik.http.routers.chat-new-sdk-prompt.middlewares=chat-new-sdk-prompt-strip"
+      - "traefik.http.services.chat-new-sdk-prompt.loadbalancer.server.port=80"
+      - "traefik.http.routers.chat-new-sdk-prompt.service=chat-new-sdk-prompt"
+
+networks:
+  agenta-network:
+    external: true

services/chat-new-sdk.rest

@@ -8,6 +8,35 @@
 ### Health Check
 GET {{baseUrl}}/{{service}}/health HTTP/1.1
 
+### Test chat-new-sdk
+POST {{baseUrl}}/{{service}}/chat HTTP/1.1
+Content-Type: application/json
+
+{
+    "inputs": {
+        "message": "What is the capital of France?"
+    }
+}
+
+### Test chat configuration
+POST {{baseUrl}}/{{service}}/configure HTTP/1.1
+Content-Type: application/json
+
+{
+    "model": "gpt-3.5-turbo",
+    "temperature": 0.7,
+    "max_tokens": 100,
+    "prompt": {
+        "messages": [
+            {
+                "role": "system",
+                "content": "You are a helpful assistant specializing in geography."
+            }
+        ],
+        "template_format": "fstring"
+    }
+}
+
 ### Generate Chat Response
 POST {{baseUrl}}/{{service}}/generate HTTP/1.1
 Content-Type: application/json

services/chat-new-sdk/_app.py

@@ -0,0 +1,65 @@
+from typing import Annotated, Any, Dict, List
+
+import agenta as ag
+from agenta.sdk.assets import supported_llm_models
+from pydantic import BaseModel, Field
+import os
+# Import mock if MOCK_LLM environment variable is set
+if os.getenv("MOCK_LLM", True):
+    from mock_litellm import MockLiteLLM
+
+    litellm = MockLiteLLM()
+else:
+    import litellm
+
+    litellm.drop_params = True
+    litellm.callbacks = [ag.callbacks.litellm_handler()]
+
+SYSTEM_PROMPT = "You have expertise in offering technical ideas to startups."
+
+ag.init()
+
+
+class MyConfig(BaseModel):
+    temperature: float = Field(default=0.2, le=1, ge=0)
+    model: Annotated[str, ag.MultipleChoice(choices=supported_llm_models)] = Field(
+        default="gpt-3.5-turbo"
+    )
+    max_tokens: int = Field(default=-1, ge=-1, le=4000)
+    prompt_system: str = Field(default=SYSTEM_PROMPT)
+
+
+@ag.instrument(spankind="llm")
+async def llm_call(messages: List[Dict[str, Any]], maxtokens):
+    config = ag.ConfigManager.get_from_route(schema=MyConfig)
+    chat_completion = await litellm.acompletion(
+        model=config.model,
+        messages=messages,
+        temperature=config.temperature,
+        max_tokens=maxtokens,
+    )
+    token_usage = chat_completion.usage.dict()
+    return {
+        "usage": token_usage,
+        "message": chat_completion.choices[0].message.content,
+        "cost": litellm.cost_calculator.completion_cost(
+            completion_response=chat_completion, model=config.model
+        ),
+    }
+
+
+@ag.route("/", config_schema=MyConfig)
+@ag.instrument()
+async def chat(inputs: ag.MessagesInput = ag.MessagesInput()) -> Dict[str, Any]:
+    config = ag.ConfigManager.get_from_route(schema=MyConfig)
+    messages = [{"role": "system", "content": config.prompt_system}] + inputs
+    max_tokens = config.max_tokens if config.max_tokens != -1 else None
+    response = await llm_call(
+        messages=messages,
+        maxtokens=max_tokens,
+    )
+    return {
+        "message": response["message"],
+        "usage": response.get("usage", None),
+        "cost": response.get("cost", None),
+    }

services/chat-new-sdk/docker-compose.yml

@@ -0,0 +1,23 @@
+services:
+  chat-new-sdk:
+    build: .
+    volumes:
+      - .:/app
+      - ../../agenta-cli:/agenta-cli
+    environment:
+      - AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED=True
+      - AGENTA_HOST=http://host.docker.internal
+    networks:
+      - agenta-network
+    labels:
+      - "traefik.http.routers.chat-new-sdk.rule=PathPrefix(`/chat-new-sdk/`)"
+      - "traefik.http.routers.chat-new-sdk.entrypoints=web"
+      - "traefik.http.middlewares.chat-new-sdk-strip.stripprefix.prefixes=/chat-new-sdk"
+      - "traefik.http.middlewares.chat-new-sdk-strip.stripprefix.forceslash=true"
+      - "traefik.http.routers.chat-new-sdk.middlewares=chat-new-sdk-strip"
+      - "traefik.http.services.chat-new-sdk.loadbalancer.server.port=80"
+      - "traefik.http.routers.chat-new-sdk.service=chat-new-sdk"
+
+networks:
+  agenta-network:
+    external: true

services/completion-new-sdk-prompt.rest

@@ -0,0 +1,37 @@
+### Test completion-new-sdk-prompt
+POST http://localhost/completion-new-sdk-prompt/generate
+Content-Type: application/json
+
+{
+    "inputs": {
+        "country": "France"
+    }
+}
+
+### Test completion configuration with prompt
+POST http://localhost/completion-new-sdk-prompt/configure
+Content-Type: application/json
+
+{
+    "model": "gpt-3.5-turbo",
+    "temperature": 0.7,
+    "max_tokens": 100,
+    "prompt": {
+        "messages": [
+            {
+                "role": "system",
+                "content": "You are a helpful assistant."
+            },
+            {
+                "role": "user",
+                "content": "What is the capital of {country}?"
+            }
+        ],
+        "template_format": "fstring",
+        "response_format": {
+            "type": "text"
+        },
+        "tools": null,
+        "tool_choice": "auto"
+    }
+}

services/completion-new-sdk-prompt/Dockerfile

@@ -0,0 +1,18 @@
+FROM python:3.10-slim
+
+ARG ROOT_PATH=/
+ENV ROOT_PATH=${ROOT_PATH}
+
+WORKDIR /app
+
+COPY . .
+
+RUN pip install --upgrade pip \
+    && pip install --no-cache-dir agenta openai python-dotenv uvicorn "litellm>=1.0,<2.0" google-generativeai
+
+# Add agenta-cli to PYTHONPATH so it can find the local agenta package
+ENV PYTHONPATH=/agenta-cli:$PYTHONPATH
+
+EXPOSE 80
+
+CMD ["./entrypoint.sh"]

services/completion-new-sdk-prompt/Dockerfile.prerelease

@@ -0,0 +1,16 @@
+FROM python:3.10-slim
+
+ARG ROOT_PATH=/
+ENV ROOT_PATH=${ROOT_PATH}
+
+WORKDIR /app
+
+COPY . .
+
+RUN pip install --upgrade pip \
+    && pip install --no-cache-dir openai python-dotenv uvicorn "litellm>=1.0,<2.0" google-generativeai \
+    && pip install --no-cache-dir --pre agenta
+
+EXPOSE 80
+
+CMD ["./entrypoint.sh"]