agenta 0.12.7__py3-none-any.whl → 0.13.0__py3-none-any.whl

agenta/sdk/tracing/llm_tracing.py

@@ -0,0 +1,220 @@
+# Stdlib Imports
+from datetime import datetime, timezone
+from typing import Optional, Dict, Any, List, Union
+
+# Own Imports
+from agenta.sdk.tracing.logger import llm_logger
+from agenta.sdk.tracing.tasks_manager import TaskQueue
+from agenta.client.backend.client import AsyncAgentaApi
+from agenta.client.backend.client import AsyncObservabilityClient
+from agenta.client.backend.types.create_span import CreateSpan, SpanKind, SpanStatusCode
+
+# Third Party Imports
+from bson.objectid import ObjectId
+
+
+class Tracing(object):
+    """Agenta llm tracing object.
+
+    Args:
+        base_url (str): The URL of the backend host
+        api_key (str): The API Key of the backend host
+        tasks_manager (TaskQueue): The tasks manager dedicated to handling asynchronous tasks
+        llm_logger (Logger): The logger associated with the LLM tracing
+        max_workers (int): The maximum number of workers to run tracing
+    """
+
+    _instance = None
+
+    def __new__(cls, *args, **kwargs):
+        if not cls._instance:
+            cls._instance = super().__new__(cls)
+        return cls._instance
+
+    def __init__(
+        self,
+        base_url: str,
+        app_id: str,
+        variant_id: str,
+        variant_name: Optional[str] = None,
+        api_key: Optional[str] = None,
+        max_workers: Optional[int] = None,
+    ):
+        self.base_url = base_url + "/api"
+        self.api_key = api_key if api_key is not None else ""
+        self.llm_logger = llm_logger
+        self.app_id = app_id
+        self.variant_id = variant_id
+        self.variant_name = variant_name
+        self.tasks_manager = TaskQueue(
+            max_workers if max_workers else 4, logger=llm_logger
+        )
+        self.active_span = CreateSpan
+        self.active_trace = CreateSpan
+        self.recording_trace_id: Union[str, None] = None
+        self.recorded_spans: List[CreateSpan] = []
+        self.tags: List[str] = []
+        self.span_dict: Dict[str, CreateSpan] = {}  # type: ignore
+
+    @property
+    def client(self) -> AsyncObservabilityClient:
+        """Initialize observability async client
+
+        Returns:
+            AsyncObservabilityClient: async client
+        """
+
+        return AsyncAgentaApi(
+            base_url=self.base_url, api_key=self.api_key, timeout=120  # type: ignore
+        ).observability
+
+    def set_span_attribute(
+        self, parent_key: Optional[str] = None, attributes: Dict[str, Any] = {}
+    ):
+        span = self.span_dict[self.active_span.id]  # type: ignore
+        for key, value in attributes.items():
+            self.set_attribute(span.attributes, key, value, parent_key)  # type: ignore
+
+    def set_attribute(
+        self,
+        span_attributes: Dict[str, Any],
+        key: str,
+        value: Any,
+        parent_key: Optional[str] = None,
+    ):
+        if parent_key is not None:
+            model_config = span_attributes.get(parent_key, None)
+            if not model_config:
+                span_attributes[parent_key] = {}
+            span_attributes[parent_key][key] = value
+        else:
+            span_attributes[key] = value
+
+    def set_trace_tags(self, tags: List[str]):
+        self.tags.extend(tags)
+
+    def start_parent_span(
+        self, name: str, inputs: Dict[str, Any], config: Dict[str, Any], **kwargs
+    ):
+        trace_id = self._create_trace_id()
+        span_id = self._create_span_id()
+        self.llm_logger.info("Recording parent span...")
+        span = CreateSpan(
+            id=span_id,
+            app_id=self.app_id,
+            variant_id=self.variant_id,
+            variant_name=self.variant_name,
+            inputs=inputs,
+            name=name,
+            config=config,
+            environment=kwargs.get("environment"),
+            spankind=SpanKind.WORKFLOW.value,
+            status=SpanStatusCode.UNSET.value,
+            start_time=datetime.now(timezone.utc),
+        )
+        self.active_trace = span
+        self.recording_trace_id = trace_id
+        self.parent_span_id = span.id
+        self.llm_logger.info(
+            f"Recorded active_trace and setting parent_span_id: {span.id}"
+        )
+
+    def start_span(
+        self,
+        name: str,
+        spankind: str,
+        input: Dict[str, Any],
+        config: Dict[str, Any] = {},
+    ) -> CreateSpan:
+        span_id = self._create_span_id()
+        self.llm_logger.info(f"Recording {spankind} span...")
+        span = CreateSpan(
+            id=span_id,
+            inputs=input,
+            name=name,
+            app_id=self.app_id,
+            variant_id=self.variant_id,
+            variant_name=self.variant_name,
+            config=config,
+            environment=self.active_trace.environment,
+            parent_span_id=self.parent_span_id,
+            spankind=spankind.upper(),
+            attributes={},
+            status=SpanStatusCode.UNSET.value,
+            start_time=datetime.now(timezone.utc),
+        )
+
+        self.active_span = span
+        self.span_dict[span.id] = span
+        self.parent_span_id = span.id
+        self.llm_logger.info(
+            f"Recorded active_span and setting parent_span_id: {span.id}"
+        )
+        return span
+
+    def update_span_status(self, span: CreateSpan, value: str):
+        updated_span = CreateSpan(**{**span.dict(), "status": value})
+        self.active_span = updated_span
+
+    def end_span(self, outputs: Dict[str, Any], span: CreateSpan, **kwargs):
+        updated_span = CreateSpan(
+            **span.dict(),
+            end_time=datetime.now(timezone.utc),
+            outputs=[outputs["message"]],
+            cost=outputs.get("cost", None),
+            tokens=outputs.get("usage"),
+        )
+
+        # Push span to list of recorded spans
+        self.recorded_spans.append(updated_span)
+        self.llm_logger.info(
+            f"Pushed {updated_span.spankind} span {updated_span.id} to recorded spans."
+        )
+
+    def end_recording(self, outputs: Dict[str, Any], span: CreateSpan, **kwargs):
+        self.end_span(outputs=outputs, span=span, **kwargs)
+        if self.api_key == "":
+            return
+
+        self.llm_logger.info(f"Preparing to send recorded spans for processing.")
+        self.llm_logger.info(f"Recorded spans => {len(self.recorded_spans)}")
+        self.tasks_manager.add_task(
+            self.active_trace.id,
+            "trace",
+            self.client.create_traces(
+                trace=self.recording_trace_id, spans=self.recorded_spans  # type: ignore
+            ),
+            self.client,
+        )
+        self.llm_logger.info(
+            f"Tracing for {span.id} recorded successfully and sent for processing."
+        )
+        self._clear_recorded_spans()
+
+    def _create_trace_id(self) -> str:
+        """Creates a unique mongo id for the trace object.
+
+        Returns:
+            str: stringify oid of the trace
+        """
+
+        return str(ObjectId())
+
+    def _create_span_id(self) -> str:
+        """Creates a unique mongo id for the span object.
+
+        Returns:
+            str: stringify oid of the span
+        """
+
+        return str(ObjectId())
+
+    def _clear_recorded_spans(self) -> None:
+        """
+        Clear the list of recorded spans to prepare for next batch processing.
+        """
+
+        self.recorded_spans = []
+        self.llm_logger.info(
+            f"Cleared all recorded spans from batch: {self.recorded_spans}"
+        )

agenta/sdk/tracing/logger.py

@@ -0,0 +1,19 @@
+import logging
+
+
+class LLMLogger:
+    def __init__(self, name="LLMLogger", level=logging.INFO):
+        self.logger = logging.getLogger(name)
+        self.logger.setLevel(level)
+
+        # Add a stream logger to view the logs in the console
+        console_handler = logging.StreamHandler()
+        self.logger.addHandler(console_handler)
+
+    @property
+    def log(self) -> logging.Logger:
+        return self.logger
+
+
+# Initialize llm logger
+llm_logger = LLMLogger().log

agenta/sdk/tracing/tasks_manager.py

@@ -0,0 +1,130 @@
+# Stdlib Imports
+import queue
+import asyncio
+from logging import Logger
+from typing import Coroutine, Optional, Union
+from concurrent.futures import ThreadPoolExecutor
+
+# Own Imports
+from agenta.client.backend.types.error import Error
+from agenta.client.backend.client import AsyncObservabilityClient
+
+
+class AsyncTask(object):
+    """Wraps a coroutine (an asynchronous function defined with async def).
+
+    Args:
+        coroutine (Coroutine): asynchronous function
+    """
+
+    def __init__(
+        self,
+        coroutine_id: str,
+        coroutine_type: str,
+        coroutine: Coroutine,
+        client: AsyncObservabilityClient,
+    ):
+        self.coroutine_id = coroutine_id
+        self.coroutine_type = coroutine_type
+        self.coroutine = coroutine
+        self.task: Optional[asyncio.Task] = None
+        self.client = client
+
+    async def run(self) -> Union[asyncio.Task, Error]:
+        """Creates an asyncio Task from the coroutine and starts it
+
+        Returns:
+            Task: asyncio task
+        """
+
+        try:
+            self.task = asyncio.create_task(self.coroutine)
+        except Exception as exc:
+            return Error(message="error running task", stacktrace=str(exc))
+        return await self.task
+
+    def cancel(self):
+        """
+        Cancels running asyncio Task.
+        """
+
+        if self.task:
+            self.task.cancel()
+
+
+class TaskQueue(object):
+    """Stores a list of AsyncTask instances.
+
+    Args:
+        tasks (List[AsyncTasks]): list of async task instances
+
+    Example Usage:
+        ```python
+        queue = TaskQueue()
+        queue.add_task(long_running_task(1))
+        queue.add_task(long_running_task(2))
+        ```
+    """
+
+    def __init__(self, num_workers: int, logger: Logger):
+        self.tasks = queue.Queue()  # type: ignore
+        self._logger = logger
+        self._thread_pool = ThreadPoolExecutor(max_workers=num_workers)
+
+    def add_task(
+        self,
+        coroutine_id: str,
+        coroutine_type: str,
+        coroutine: Coroutine,
+        obs_client: AsyncObservabilityClient,
+    ) -> AsyncTask:
+        """Adds a new task to be executed.
+
+        Args:
+            coroutine_id (str): The Id of the coroutine
+            coroutine_type (str): The type of coroutine
+            coroutine (Coroutine): async task
+            obs_client (AsyncObservabilityClient): The async observability client
+
+        Returns:
+            AsyncTask: task to be executed
+        """
+
+        task = AsyncTask(coroutine_id, coroutine_type, coroutine, obs_client)
+        self.tasks.put(task)
+        return self._worker()
+
+    def _worker(self):
+        """
+        Runs task gotten from the queue in a thread pool.
+        """
+
+        while True:
+            task: AsyncTask = self.tasks.get()  # type: ignore
+            try:
+                future = self._thread_pool.submit(asyncio.run, task.run())
+                future.result()
+            except Exception as exc:
+                self._logger.error(f"Error running task: {str(exc)}")
+                self._logger.error(
+                    f"Recording trace {task.coroutine_type} status to ERROR."
+                )
+                break
+            finally:
+                self.tasks.task_done()
+                break
+
+    def _get_size(self) -> int:
+        """Returns the approximate number of items in the queue."""
+
+        return self.tasks.qsize()
+
+    def flush(self) -> None:
+        """Clears all items from the queue."""
+
+        q_size = self._get_size()
+        self._logger.info("Flushing queue...")
+        with self.tasks.mutex:  # acts as a lock to ensure that only one thread can access the queue
+            self.tasks.join()
+            self._logger.info(f"Queue with {q_size} items flushed successfully")
+            return

{agenta-0.12.7.dist-info → agenta-0.13.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: agenta
-Version: 0.12.7
+Version: 0.13.0
 Summary: The SDK for agenta is an open-source LLMOps platform.
 Home-page: https://agenta.ai
 Keywords: LLMOps,LLM,evaluation,prompt engineering
@@ -23,6 +23,7 @@ Requires-Dist: importlib-metadata (>=6.7.0,<7.0.0)
 Requires-Dist: ipdb (>=0.13)
 Requires-Dist: posthog (>=3.1.0,<4.0.0)
 Requires-Dist: pydantic (==1.10.13)
+Requires-Dist: pymongo (>=4.6.3,<5.0.0)
 Requires-Dist: python-dotenv (>=1.0.0,<2.0.0)
 Requires-Dist: python-multipart (>=0.0.6,<0.0.7)
 Requires-Dist: questionary (>=1.10.0,<2.0.0)
@@ -35,8 +36,8 @@ Description-Content-Type: text/markdown
   <a href="https://agenta.ai/">
     <div  align="center" >
       <picture >
-        <source media="(prefers-color-scheme: dark)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/a356f263-6f5e-43df-8b58-4c183b8d8878"  >
-        <source media="(prefers-color-scheme: light)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/68e055d4-d7b8-4943-992f-761558c64253"  >
+        <source width="500" media="(prefers-color-scheme: dark)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/cdddf5ad-2352-4920-b1d9-ae7f8d9d7735"  >
+        <source width="500" media="(prefers-color-scheme: light)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/ab75cbac-b807-496f-aab3-57463a33f726"  >
         <img alt="Shows the logo of agenta" src="https://github.com/Agenta-AI/agenta/assets/4510758/68e055d4-d7b8-4943-992f-761558c64253" >
       </picture>
     </div>
@@ -47,8 +48,8 @@ Description-Content-Type: text/markdown
     <a href="https://docs.agenta.ai/">Documentation</a>
   </h4>
 <div align="center">
-  <strong>Quickly iterate, debug, and evaluate your LLM apps</strong><br />
-  The open-source LLMOps platform for prompt-engineering, evaluation, human feedback, and deployment of complex LLM apps.
+  <strong> <h2> Collaborate on prompts, evaluate, and deploy LLM applications with confidence </h2></strong>
+  The open-source LLM developer platform for prompt-engineering, evaluation, human feedback, and deployment of complex LLM apps.
 </div>
 </br>
 <p align="center">
@@ -67,7 +68,6 @@ Description-Content-Type: text/markdown
   <a href="https://pypi.org/project/agenta/">
     <img src="https://img.shields.io/pypi/dm/agenta" alt="PyPI - Downloads">
   </a>
-
 </br>
 </p>
 
@@ -89,7 +89,7 @@ Description-Content-Type: text/markdown
 
 <a href="https://cloud.agenta.ai">
   <picture>
-  <img src="https://github.com/Agenta-AI/agenta/assets/4510758/a3024fac-2945-4208-ae12-4cc51ecfc970" />
+  <img src="https://github.com/Agenta-AI/agenta/assets/4510758/d622f268-c295-42d4-a43c-548b611ff6f2" />
   </picture>
 </a>
 
@@ -99,8 +99,8 @@ Description-Content-Type: text/markdown
   <br />
       <div  align="center" >
       <picture >
-        <source media="(prefers-color-scheme: dark)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/cf6d4713-4558-4c6c-9e1b-ee4eab261f4c"  >
-        <source media="(prefers-color-scheme: light)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/ae9cf11f-8ef9-4b67-98c7-4a40341fa87a"  >
+        <source width="800" media="(prefers-color-scheme: dark)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/cf6d4713-4558-4c6c-9e1b-ee4eab261f4c"  >
+        <source width="800" media="(prefers-color-scheme: light)" srcset="https://github.com/Agenta-AI/agenta/assets/4510758/ae9cf11f-8ef9-4b67-98c7-4a40341fa87a"  >
         <img alt="Mockup agenta" src="https://github.com/Agenta-AI/agenta/assets/4510758/ae9cf11f-8ef9-4b67-98c7-4a40341fa87a" >
       </picture>
     </div>
@@ -112,126 +112,75 @@ Description-Content-Type: text/markdown
 ---
 
 <h3 align="center">
-  <a href="#ℹ️-about"><b>About</b></a> &bull;
   <a href="#quick-start"><b>Quick Start</b></a> &bull;
-  <a href="https://docs.agenta.ai/installation"><b>Installation</b></a> &bull;
   <a href="#features"><b>Features</b></a> &bull;
   <a href="https://docs.agenta.ai"><b>Documentation</b></a> &bull;
   <a href="#enterprise-support"><b>Enterprise</b></a> &bull;
-  <a href="https://join.slack.com/t/agenta-hq/shared_invite/zt-1zsafop5i-Y7~ZySbhRZvKVPV5DO_7IA"><b>Community</b></a> &bull;
+  <a href="https://github.com/orgs/Agenta-AI/projects/13"><b>Roadmap</b></a> &bull;
+  <a href="https://join.slack.com/t/agenta-hq/shared_invite/zt-1zsafop5i-Y7~ZySbhRZvKVPV5DO_7IA"><b>Join Our Slack</b></a> &bull;
   <a href="#contributing"><b>Contributing</b></a>
 </h3>
 
 ---
 
-# ℹ️ About
+# ⭐️ Why Agenta?
+
+Agenta is an end-to-end LLM developer platform. It provides the tools for **prompt engineering and management**,  ⚖️ **evaluation**, **human annotation**, and :rocket: **deployment**. All without imposing any restrictions on your choice of framework, library, or model. 
+
+Agenta allows developers and product teams to collaborate in building production-grade LLM-powered applications in less time. 
+
+### With Agenta, you can:
 
-Agenta is an end-to-end LLMOps platform. It provides the tools for **prompt engineering and management**,  ⚖️ **evaluation**, and :rocket: **deployment**. All without imposing any restrictions on your choice of framework, library, or model. 
+- [🧪 **Experiment** and **compare** prompts](https://docs.agenta.ai/basic_guides/prompt_engineering) on [any LLM workflow](https://docs.agenta.ai/advanced_guides/custom_applications) (chain-of-prompts, Retrieval Augmented Generation (RAG), LLM agents...) 
+- ✍️ Collect and [**annotate golden test sets**](https://docs.agenta.ai/basic_guides/test_sets) for evaluation
+- 📈 [**Evaluate** your application](https://docs.agenta.ai/basic_guides/automatic_evaluation) with pre-existing or [**custom evaluators**](https://docs.agenta.ai/advanced_guides/using_custom_evaluators)
+- [🔍 **Annotate** and **A/B test**](https://docs.agenta.aibasic_guides/human_evaluation) your applications with **human feedback**
+- [🤝 **Collaborate with product teams**](https://docs.agenta.ai/basic_guides/team_management) for prompt engineering and evaluation
+- [🚀 **Deploy your application**](https://docs.agenta.ai/basic_guides/deployment) in one-click in the UI,  through CLI, or through github workflows. 
 
-Agenta allows developers and product teams to collaborate and build robust AI applications in less time. 
+### Works with any LLM app workflow
 
-## 🔨 How does it work?
+Agenta enables prompt engineering and evaluation on any LLM app architecture:
+- Chain of prompts
+- RAG
+- Agents
+- ...
 
-| Using an LLM App Template (For Non-Technical Users)  | Starting from Code |
-| ------------- | ------------- |
-|1. [Create an application using a pre-built template from our UI](https://cloud.agenta.ai?utm_source=github&utm_medium=readme&utm_campaign=github)<br />2. Access a playground where you can test and compare different prompts and configurations side-by-side.<br /> 3. Systematically evaluate your application using pre-built or custom evaluators.<br /> 4. Deploy the application to production with one click. |1. [Add a few lines to any LLM application code to automatically create a playground for it](https://docs.agenta.ai/developer_guides/tutorials/first-app-with-langchain) <br />2. Experiment with prompts and configurations, and compare them side-by-side in the playground. <br />3. Systematically evaluate your application using pre-built or custom evaluators. <br />4. Deploy the application to production with one click. |
+It works with any framework such as [Langchain](https://langchain.com), [LlamaIndex](https://www.llamaindex.ai/) and any LLM provider (openAI, Cohere, Mistral). 
 
-<br /><br />
+[Jump here to see how to use your own custom application with agenta](/advanced_guides/custom_applications)
 
 # Quick Start
 
-### [Try the cloud version](https://cloud.agenta.ai?utm_source=github&utm_medium=readme&utm_campaign=github)
-### [Create your first application in one-minute](https://docs.agenta.ai/getting_started/getting-started-ui)
-### [Create an application using Langchain](https://docs.agenta.ai/developer_guides/tutorials/first-app-with-langchain)
+### [Get started for free](https://cloud.agenta.ai?utm_source=github&utm_medium=readme&utm_campaign=github)
+### [Explore the Docs](https://docs.agenta.ai)
+### [Create your first application in one-minute](https://docs.agenta.ai/quickstart/getting-started-ui)
+### [Create an application using Langchain](https://docs.agenta.ai/tutorials/first-app-with-langchain)
 ### [Self-host agenta](https://docs.agenta.ai/self-host/host-locally)
-### [Read the Documentation](https://docs.agenta.ai)
 ### [Check the Cookbook](https://docs.agenta.ai/cookbook)
 
 # Features
 
-<h3>Playground 🪄 <br/></h3>
-  With just a few lines of code, define the parameters and prompts you wish to experiment with. You and your team can quickly experiment and test new variants on the web UI. <br/>
-
-https://github.com/Agenta-AI/agenta/assets/4510758/8b736d2b-7c61-414c-b534-d95efc69134c
-
-<h3>Version Evaluation 📊 <br/></h3>
-Define test sets, then evaluate manually or programmatically your different variants.<br/>
-
-![](https://github.com/Agenta-AI/agenta/assets/4510758/b1de455d-7e0a-48d6-8497-39ba641600f0)
-
-
-<h3>API Deployment 🚀<br/></h3>
-When you are ready, deploy your LLM applications as APIs in one click.<br/>
-
-![](https://github.com/Agenta-AI/agenta/blob/main/docs/images/endpoint.gif)
-
-## Why choose Agenta for building LLM-apps?
-
-- 🔨 **Build quickly**: You need to iterate many times on different architectures and prompts to bring apps to production. We streamline this process and allow you to do this in days instead of weeks.
-- 🏗️ **Build robust apps and reduce hallucination**: We provide you with the tools to systematically and easily evaluate your application to make sure you only serve robust apps to production.
-- 👨‍💻 **Developer-centric**: We cater to complex LLM-apps and pipelines that require more than one simple prompt. We allow you to experiment and iterate on apps that have complex integration, business logic, and many prompts.
-- 🌐 **Solution-Agnostic**: You have the freedom to use any libraries and models, be it Langchain, llma_index, or a custom-written alternative.
-- 🔒 **Privacy-First**: We respect your privacy and do not proxy your data through third-party services. The platform and the data are hosted on your infrastructure.
-
-## How Agenta works:
-
-**1. Write your LLM-app code**
-
-Write the code using any framework, library, or model you want. Add the `agenta.post` decorator and put the inputs and parameters in the function call just like in this example:
-
-_Example simple application that generates baby names:_
-
-```python
-import agenta as ag
-from langchain.chains import LLMChain
-from langchain.llms import OpenAI
-from langchain.prompts import PromptTemplate
-
-default_prompt = "Give me five cool names for a baby from {country} with this gender {gender}!!!!"
-ag.init()
-ag.config(prompt_template=ag.TextParam(default_prompt),
-          temperature=ag.FloatParam(0.9))
-
-@ag.entrypoint
-def generate(
-    country: str,
-    gender: str,
-) -> str:
-    llm = OpenAI(temperature=ag.config.temperature)
-    prompt = PromptTemplate(
-        input_variables=["country", "gender"],
-        template=ag.config.prompt_template,
-    )
-    chain = LLMChain(llm=llm, prompt=prompt)
-    output = chain.run(country=country, gender=gender)
-
-    return output
-```
-
-**2.Deploy your app using the Agenta CLI**
-
-<img width="650" alt="Screenshot 2023-06-19 at 15 58 34" src="https://github.com/Agenta-AI/agenta/assets/4510758/eede3e78-0fe1-42a0-ad4e-d880ddb10bf0">
-
-**3. Go to agenta at http://localhost**
-
-Now your team can 🔄 iterate, 🧪 experiment, and ⚖️ evaluate different versions of your app (with your code!) in the web platform.</summary>
-<br/>
-
-<img width="900" alt="Screenshot 2023-06-25 at 21 08 53" src="https://github.com/Agenta-AI/agenta/assets/57623556/7e07a988-a36a-4fb5-99dd-9cc13a678434">
 
+| Playground | Evaluation |
+| ------- | ------- |
+| Compare and version prompts for any LLM app, from single prompt to agents. <br/> <video src="https://github.com/Agenta-AI/agenta/assets/4510758/8b736d2b-7c61-414c-b534-d95efc69134c" controls="controls" style="max-width:100%;"> | Define test sets, then evaluate manually or programmatically your different variants.<br/> <video src="https://github.com/Agenta-AI/agenta/assets/4510758/8c6997c6-da87-46ad-a81f-e15e277263d2" controls="controls" style="max-width:100%;">|
+| Human annotation | Deployment |
+| Use Human annotator to A/B test and score your LLM apps. <br/>  <img width="750" alt="Screenshot 2024-01-28 at 12 57 46" src="https://github.com/Agenta-AI/agenta/assets/4510758/bf62a697-bf19-4ba9-850e-742fbfb75424"> | When you are ready, deploy your LLM applications as APIs in one click.<br/>![](https://github.com/Agenta-AI/agenta/blob/main/docs/images/endpoint.gif) |
 
 # Enterprise Support
 Contact us here for enterprise support and early access to agenta self-managed enterprise with Kubernetes support. <br/><br/>
 <a href="https://cal.com/mahmoud-mabrouk-ogzgey/demo"><img src="https://cal.com/book-with-cal-dark.svg" alt="Book us"></a>
 
 # Disabling Anonymized Tracking
+By default, Agenta automatically reports anonymized basic usage statistics. This helps us understand how Agenta is used and track its overall usage and growth. This data does not include any sensitive information.
 
-To disable anonymized telemetry, set the following environment variable:
+To disable anonymized telemetry, follow these steps:
 
 - For web: Set `TELEMETRY_TRACKING_ENABLED` to `false` in your `agenta-web/.env` file.
 - For CLI: Set `telemetry_tracking_enabled` to `false` in your `~/.agenta/config.toml` file.
 
-After making this change, restart agenta compose.
+After making this change, restart Agenta Compose.
 
 # Contributing
 
@@ -244,7 +193,7 @@ Check out our [Contributing Guide](https://docs.agenta.ai/contributing/getting-s
 ## Contributors ✨
 
 <!-- ALL-CONTRIBUTORS-BADGE:START - Do not remove or modify this section -->
-[![All Contributors](https://img.shields.io/badge/all_contributors-39-orange.svg?style=flat-square)](#contributors-)
+[![All Contributors](https://img.shields.io/badge/all_contributors-41-orange.svg?style=flat-square)](#contributors-)
 <!-- ALL-CONTRIBUTORS-BADGE:END -->
 
 Thanks goes to these wonderful people ([emoji key](https://allcontributors.org/docs/en/emoji-key)):
@@ -304,6 +253,8 @@ Thanks goes to these wonderful people ([emoji key](https://allcontributors.org/d
       <td align="center" valign="top" width="14.28%"><a href="https://github.com/brockWith"><img src="https://avatars.githubusercontent.com/u/105627491?v=4?s=100" width="100px;" alt="brockWith"/><br /><sub><b>brockWith</b></sub></a><br /><a href="https://github.com/Agenta-AI/agenta/commits?author=brockWith" title="Code">💻</a></td>
       <td align="center" valign="top" width="14.28%"><a href="http://denniszelada.wordpress.com/"><img src="https://avatars.githubusercontent.com/u/219311?v=4?s=100" width="100px;" alt="Dennis Zelada"/><br /><sub><b>Dennis Zelada</b></sub></a><br /><a href="https://github.com/Agenta-AI/agenta/commits?author=denniszelada" title="Code">💻</a></td>
       <td align="center" valign="top" width="14.28%"><a href="https://github.com/romainrbr"><img src="https://avatars.githubusercontent.com/u/10381609?v=4?s=100" width="100px;" alt="Romain Brucker"/><br /><sub><b>Romain Brucker</b></sub></a><br /><a href="https://github.com/Agenta-AI/agenta/commits?author=romainrbr" title="Code">💻</a></td>
+      <td align="center" valign="top" width="14.28%"><a href="http://heonheo.com"><img src="https://avatars.githubusercontent.com/u/76820291?v=4?s=100" width="100px;" alt="Heon Heo"/><br /><sub><b>Heon Heo</b></sub></a><br /><a href="https://github.com/Agenta-AI/agenta/commits?author=HeonHeo23" title="Code">💻</a></td>
+      <td align="center" valign="top" width="14.28%"><a href="https://github.com/Drewski2222"><img src="https://avatars.githubusercontent.com/u/39228951?v=4?s=100" width="100px;" alt="Drew Reisner"/><br /><sub><b>Drew Reisner</b></sub></a><br /><a href="https://github.com/Agenta-AI/agenta/commits?author=Drewski2222" title="Code">💻</a></td>
     </tr>
   </tbody>
 </table>