visibe 0.1.5

AI Agent Observability Platform - Track CrewAI, LangChain, LangGraph, and more

Visibe SDK for Python

Observability for AI agents. Track costs, performance, and errors across your entire AI stack — whether you're using CrewAI, LangChain, LangGraph, AutoGen, or direct OpenAI calls.

---

📦 Getting Started

1. Create an account

Sign up at app.visibe.ai and create a project.

2. Get an API key

In your project, go to Settings → API Keys and generate a new key. It will look like sk_live_....

3. Install the SDK

pip install visibe

4. Set your API key

export VISIBE_API_KEY=sk_live_your_api_key_here

Or pass it directly in code:

visibe.init(api_key="sk_live_your_api_key_here")

Or load it from a .env file using python-dotenv:

pip install python-dotenv

from dotenv import load_dotenv
load_dotenv()  # loads VISIBE_API_KEY from .env

import visibe
visibe.init()

5. Instrument your app

import visibe

visibe.init()

That's it. Every OpenAI, LangChain, LangGraph, CrewAI, AutoGen, and Bedrock client created after this call is automatically traced — no other code changes needed.

---

🧩 Integrations

Framework	Auto (visibe.init())	Manual
OpenAI	✅	✅
LangChain	✅	✅
LangGraph	✅	✅
CrewAI	✅	✅
AutoGen	✅	✅
AWS Bedrock	✅	✅

Also works with OpenAI-compatible providers: Azure OpenAI, Groq, Together.ai, DeepSeek, and others.

OpenAI

import visibe
from openai import OpenAI

visibe.init()

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello!"}]
)
# Automatically traced — cost, tokens, duration, and content captured.

LangChain / LangGraph

import visibe
from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

visibe.init()

llm = ChatOpenAI(model="gpt-4o-mini")
graph = create_react_agent(llm, tools)

result = graph.invoke({"messages": [("user", "Your prompt here")]})
# Automatically traced — agent steps, LLM calls, and tool calls captured.

Dynamic pipe chains (prompt | llm | parser) are also automatically traced. Nested sub-graphs are tracked with hierarchical agent names.

CrewAI

import visibe
from crewai import Agent, Task, Crew

visibe.init()

architect = Agent(role="Plot Architect", goal="Design mystery plots", backstory="...")
designer = Agent(role="Character Designer", goal="Create characters", backstory="...")
narrator = Agent(role="Narrator", goal="Write the story", backstory="...")

task1 = Task(description="Create a plot outline", agent=architect, expected_output="...")
task2 = Task(description="Design characters", agent=designer, expected_output="...", context=[task1])
task3 = Task(description="Write the story", agent=narrator, expected_output="...", context=[task1, task2])

crew = Crew(agents=[architect, designer, narrator], tasks=[task1, task2, task3])
result = crew.kickoff()
# Single trace with all agents, LLM calls, and per-task cost breakdown.

Training and testing runs (crew.train(), crew.test()) are traced too.

AutoGen

import visibe
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_agentchat.agents import AssistantAgent

visibe.init()

model_client = OpenAIChatCompletionClient(model="gpt-4o-mini")
assistant = AssistantAgent("assistant", model_client=model_client)
result = await assistant.run(task="Help me with this task")
# Automatically traced.

AWS Bedrock

import visibe
import boto3

visibe.init()

bedrock = boto3.client("bedrock-runtime", region_name="us-east-1")
response = bedrock.converse(
    modelId="anthropic.claude-3-haiku-20240307-v1:0",
    messages=[{"role": "user", "content": [{"text": "Hello!"}]}]
)
# Automatically traced.

Supports converse, converse_stream, invoke_model, and invoke_model_with_response_stream. Works with all models available via Bedrock — Claude, Nova, Llama, Mistral, and more.

---

⚙️ Configuration

import visibe

visibe.init(
    api_key="sk_live_abc123",       # or set VISIBE_API_KEY env var
    frameworks=["openai", "langgraph"],  # limit to specific frameworks
    content_limit=500,              # max chars for LLM content in traces
    debug=True,                     # enable debug logging
)

Environment Variables

Variable	Description	Default
VISIBE_API_KEY	Your API key (required)	—
VISIBE_API_URL	Override API endpoint	https://api.visibe.ai
VISIBE_AUTO_INSTRUMENT	Comma-separated frameworks to auto-instrument	All detected
VISIBE_CONTENT_LIMIT	Max chars for LLM/tool content in spans	1000
VISIBE_DEBUG	Enable debug logging (1 to enable)	0

---

📊 What Gets Tracked

Metric	Description
Cost	Total spend + per-agent and per-task cost breakdown
Tokens	Input/output tokens per LLM call
Duration	Total time + time per step
Tools	Which tools were used, duration, success/failure
Errors	When and where things failed
Spans	Full execution timeline with LLM calls, tool calls, and agent events

---

🔧 Manual Instrumentation

For cases where you need explicit control — instrumenting a specific client, grouping calls into a named trace, or using Visibe without init().

Instrument a specific client

from visibe import Visibe

tracer = Visibe(api_key="sk_live_abc123")
tracer.instrument(graph, name="my-agent")

result = graph.invoke({"messages": [("user", "Hello")]})

Group multiple calls into one trace

from visibe import Visibe

tracer = Visibe()

with tracer.track(client, name="my-conversation"):
    r1 = client.chat.completions.create(model="gpt-4o-mini", messages=[...])
    r2 = client.chat.completions.create(model="gpt-4o-mini", messages=[...])
# Both calls sent as one grouped trace.

Remove instrumentation

tracer.uninstrument(client)

# Or use as a context manager for automatic cleanup:
with tracer.instrument(graph, name="my-agent"):
    graph.invoke(...)
# Instrumentation removed automatically on exit.

---

📚 Documentation

• OpenAI integration
• LangChain integration
• CrewAI integration
• AutoGen integration
• AWS Bedrock integration

---

🔗 Resources

• visibe.ai — Product website
• app.visibe.ai — Dashboard (sign up, manage API keys, view traces)
• PyPI Package — Latest version on PyPI

---

📃 License

MIT — see LICENSE for details.