bmasterai 0.2.4

A comprehensive Python framework for building multi-agent AI systems with advanced logging, monitoring, and integrations

BMasterAI

Production-ready AI agent monitoring, logging, and observability for Python. Drop-in telemetry for agents built on Claude, Gemini, LangGraph, or any LLM stack.

from bmasterai.logging import get_logger, EventType
from bmasterai.monitoring import get_monitor

logger = get_logger("my-agent")
monitor = get_monitor()

logger.log_event(EventType.TASK_START, "Agent started", {"task": "summarize"})
monitor.record_metric("tokens_used", 1240)

---

Examples

Real-world agents you can clone and run. Most recent first.

---

2026

Ollama Crossword Agent — Hybrid Vision + Constraint Solver NEW

March 2026

A hybrid crossword-solving agent that combines qwen2.5vl:7b (local vision model via Ollama) for reading clues and proposing answers, Playwright for deterministic browser control, and a Python constraint engine that only commits letters when crossing ACROSS and DOWN answers agree. Runs 100% locally — no API keys required.

Stack: Ollama (qwen2.5vl:7b), Playwright, BMasterAI

What it demonstrates:

• Hybrid LLM + code architecture: model proposes, code enforces — reliable solves without hallucination drift
• Crossing-constraint engine: cells committed only when all intersecting answers agree on the same letter
• Local vision inference via Ollama — screenshot → clue extraction → answer proposal in one pipeline
• Full BMasterAI instrumentation on every vision call, browser action, constraint decision, and retry
• --demo mode works offline without Ollama or a browser for easy local testing

ollama pull qwen2.5vl:7b
pip install -r requirements.txt && playwright install chromium
python main.py --demo   # no browser or Ollama needed
python main.py          # live NYT Mini Crossword

---

Gemini Web + Computer Agent — Native Function-Calling Loop

March 2026

A bare-metal Gemini function-calling agent combining web search (Tavily) and computer use (screenshot/click/type/key/scroll) — no LangGraph, no framework, just the Google GenAI SDK — fully instrumented with BMasterAI logging and telemetry. Cross-platform: works on Linux (xdotool + scrot) and macOS (cliclick + screencapture).

Stack: Gemini (Google GenAI SDK), Tavily, xdotool/cliclick, BMasterAI

What it demonstrates:

• The raw Gemini function_call / function_response message cycle — the core loop behind every Gemini agent
• Multimodal tool results: screenshots sent back to Gemini as image parts so it can see the screen
• BMasterAI telemetry on every LLM call, tool dispatch, decision point, and error path
• Structured JSONL telemetry at logs/agent.jsonl — pipe to any analytics tool

pip install -r requirements.txt
cp .env.example .env  # add GEMINI_API_KEY + TAVILY_API_KEY
python main.py "Search for today's top AI news, open a browser to the first result, take a screenshot, and summarize what you see."

---

Claude Web + Computer Agent — Native Tool-Use Loop

March 2026

A bare-metal Claude tool-use agent combining web search (Tavily) and computer use (screenshot/click/type/key/scroll) — no LangGraph, no framework, just the Anthropic SDK — fully instrumented with BMasterAI logging and telemetry. The foundational pattern that every Claude agent is built on.

Stack: Claude (Anthropic), Tavily, xdotool + scrot, BMasterAI

What it demonstrates:

• The raw Anthropic tool_use / tool_result message cycle — the core loop behind every Claude agent
• Multimodal tool results: screenshots sent back to Claude as image blocks so it can see the screen
• BMasterAI telemetry on every LLM call, tool dispatch, decision point, and error path
• Structured JSONL telemetry at logs/agent.jsonl — pipe to any analytics tool

pip install -r requirements.txt
cp .env.example .env  # add ANTHROPIC_API_KEY + TAVILY_API_KEY
python main.py "Search for today's top AI news, open a browser to the first result, take a screenshot, and summarize what you see."

---

Deep Research Agent — LangGraph + BMasterAI Telemetry

March 2026

A multi-step web research agent built with LangGraph and fully instrumented with BMasterAI logging and telemetry. Inspired by langchain-ai/deepagents. Give it any research question and it plans, searches, analyzes, reflects on quality, and synthesizes a structured report — automatically looping back for more research if gaps are found.

Stack: LangGraph, Claude (Anthropic), Tavily, BMasterAI

What it demonstrates:

• Multi-node LangGraph pipeline with a conditional reflection loop (planner → search → analyze → reflect → synthesize)
• Quality-gated research: reflector scores findings 1–10, loops back for follow-up searches when score < 7 (max 2 loops)
• BMasterAI on every step: track_agent_start/stop, track_llm_call, track_task_duration, log_event(TOOL_USE), log_reasoning_chain, log_event(DECISION_POINT)
• Structured JSONL telemetry at logs/research.jsonl — pipe to any analytics tool

pip install -r requirements.txt
cp .env.example .env  # add ANTHROPIC_API_KEY + TAVILY_API_KEY
python main.py "What is the current state of multi-agent AI systems in 2026?"

---

Viral YouTube Short Generator — LangGraph

March 2026

A four-agent LangGraph pipeline that researches trending topics and generates complete viral YouTube Short production packages — title, hook, 45-60 second script, tags, and thumbnail concept — with a quality gate that retries automatically if the output doesn't meet bar.

Stack: LangGraph, Claude (Anthropic), Tavily, BMasterAI

What it demonstrates:

• Four specialist agents in sequence: Trend Researcher → Hook Writer → Script Writer → Title & Tags
• Quality gate node with automatic retry (max 2 loops) using LangGraph conditional edges
• BMasterAI structured logging on every agent call: configure_logging, track_agent_start/stop, track_llm_call, log_event(EventType.*)
• Shared VideoState TypedDict flowing through all nodes — clean state handoff pattern

pip install -r requirements.txt
cp .env.example .env  # add ANTHROPIC_API_KEY + TAVILY_API_KEY
python main.py "AI agents taking over software engineering"

---

A2A Real Estate Multi-Agent — AgentCore Edition

March 2026

A BMasterAI adaptation of the AWS Labs A2A Real Estate sample. Three Strands agents — Property Search, Property Booking, and a Coordinator — communicate over the A2A (Agent-to-Agent) protocol, with every tool call and A2A hop instrumented via BMasterAI structured telemetry.

Stack: AWS Bedrock AgentCore, Strands, A2A Protocol, OAuth 2.0 (Cognito), BMasterAI

What it demonstrates:

• Multi-agent orchestration with the A2A protocol: coordinator delegates to specialized sub-agents at runtime
• BMasterAI telemetry replacing custom loggers: TOOL_USE, TASK_COMPLETE, TASK_ERROR on every operation across all three agents
• Bearer token forwarding from AgentCore Runtime context to sub-agent A2A calls
• Local REPL mode for development + BedrockAgentCoreApp path for serverless deployment

# Start all three agents locally + interactive REPL
pip install -r realestate_coordinator/requirements.txt
python run_local.py

---

AgentCore Memory Agent + BMasterAI Telemetry

February 2026

A Telegram bot with persistent memory built on AWS Bedrock AgentCore — fully instrumented with BMasterAI structured telemetry. The agent remembers past conversations, learns user preferences across sessions, and can execute bash commands, search the web, and send files. No Mac mini, no local server — just AWS and a Telegram message.

Stack: AWS Bedrock AgentCore, Strands, Claude (Bedrock), DynamoDB, Lambda, API Gateway, Cedar policies, BMasterAI

What it demonstrates:

• Three-strategy persistent memory (user preferences, session summaries, semantic facts) via AgentCore Memory
• Serverless Telegram webhook with session lifecycle management
• Cedar policy enforcement at the AgentCore Gateway boundary
• BMasterAI telemetry on every agent lifecycle event, memory retrieval, tool invocation, and error path — output to console, flat log, and structured JSONL

pip install -r requirements.txt
# Deploy to AgentCore Runtime:
./scripts/deploy.sh
# Set Telegram webhook:
./scripts/setup_telegram_webhook.sh

---

Amazon Bedrock AgentCore — Cost Optimization Agent

February 2026

A Strands agent that monitors AWS spend, detects anomalies, forecasts costs, and analyzes service-level breakdowns — with BMasterAI structured telemetry logged on every agent action. Inspired by the awslabs/amazon-bedrock-agentcore-samples reference implementation.

Stack: AWS Bedrock AgentCore, Strands, AWS Cost Explorer, AWS Budgets, Cost Anomaly Detection, BMasterAI

What it demonstrates:

• Cost anomaly detection, budget monitoring, and ML-based cost forecasting via Strands tools
• Full BMasterAI instrumentation: AGENT_START, TASK_START, LLM_CALL, TOOL_USE, TASK_COMPLETE on every operation
• Structured JSONL output ready for CloudWatch Insights, Datadog, or any log aggregator
• AgentCore Runtime deployment with bedrock-agentcore-starter-toolkit

pip install -r requirements.txt
python agent.py

---

WebMCP + GCP Agent Runtime

February 2026

An AI agent running on GCP Cloud Run that controls a website by calling its browser-native WebMCP tools via a Playwright bridge. The agent uses Gemini to complete shopping tasks by discovering and calling JavaScript tools registered in the browser via navigator.modelContext — instrumented end-to-end with BMasterAI.

Stack: FastAPI, Playwright, Vertex AI (Gemini 2.0 Flash), GCP Cloud Run, Docker Compose

What it demonstrates:

• WebMCP browser-native tool calling from Python via Playwright CDP bridge
• Gemini agent loop with dynamic tool discovery
• IAM-authenticated Cloud Run deployment with one-command deploy script
• BMasterAI structured logging and metrics across the full agent lifecycle

docker-compose up  # starts demo store + agent at localhost:8081
curl -X POST http://localhost:8081/run \
  -H "Content-Type: application/json" \
  -d '{"task": "Find me a laptop under $1000 and add it to the cart"}'

---

OpenClaw Telemetry Dashboard

February 2026

Real-time observability dashboard for OpenClaw AI agent sessions. Tracks LLM usage, token counts, cost estimates, tool call analytics, and session history — all in a Streamlit UI backed by BMasterAI telemetry.

Stack: Streamlit, BMasterAI, OpenClaw session logs

---

Google ADK Agent-to-Agent (A2A)

January 2026

Agent-to-Agent interaction pattern using the Google Agent Development Kit and FastMCP. A Trip Planner Agent (client) consults a Weather Agent (server) using real-time forecasts to plan trips — demonstrating multi-agent orchestration with BMasterAI monitoring at every hop.

Stack: Google ADK, FastMCP, BMasterAI

---

2025

AI LinkedIn Stress Analysis + Reasoning

August 2025

Streamlit app showing real-time Gemini reasoning transparency. Analyzes LinkedIn profiles using Tavily search and provides personalized stress reduction suggestions with full chain-of-thought visibility via BMasterAI.

Stack: Streamlit, Gemini 2.5 Pro, Tavily, BMasterAI

---

Gemini Reasoning Streamlit

August 2025

Watch Gemini 2.5 Pro think in real time. Streams chain-of-thought reasoning for complex research tasks (AI podcast influencer discovery) with Tavily web search and Firecrawl email extraction.

Stack: Streamlit, Gemini 2.5 Pro, Tavily, Firecrawl, BMasterAI

---

Streamlit + Airflow MCP Chatbot

August 2025

Natural-language interface to Apache Airflow via an MCP server. Ask questions about your DAGs, pipeline runs, and task status in plain English.

Stack: Streamlit, OpenAI, Airflow MCP, BMasterAI

---

RAG with Qdrant

August 2025

Production-ready Retrieval-Augmented Generation with async processing, intelligent caching, and real-time performance monitoring. A complete RAG reference implementation.

Stack: Qdrant, async Python, BMasterAI

---

Kubernetes Telemetry

August 2025

Kubernetes-native LLM cost analysis and observability. Wires BMasterAI metrics into OpenTelemetry, Grafana, Prometheus, Loki, and Tempo for production-grade agent monitoring at scale.

Stack: Kubernetes, Helm, OpenTelemetry, Grafana, Prometheus, BMasterAI

---

Gradio + Anthropic Claude

August 2025

Modern Gradio web interface for Claude-powered agents with BMasterAI monitoring. Clean starting point for building chat-style agent UIs.

Stack: Gradio, Claude (Anthropic), BMasterAI

---

MCP GitHub Streamlit

August 2025

Automated GitHub repo analysis and improvement suggestions using AI agents and Model Context Protocol integration.

Stack: Streamlit, MCP, BMasterAI

---

Enhanced GitHub MCP

August 2025

Advanced multi-agent system for GitHub repo analysis and automated feature implementation.

Stack: Streamlit, multi-agent, MCP, BMasterAI

---

AI Stock Research Agent

August 2025

Real-time market data, web research, and AI analysis combined into intelligent stock recommendations.

Stack: yfinance, web search, BMasterAI

---

Agno Telemetry Integration

August 2025

Full observability integration between the Agno agent framework and BMasterAI telemetry. Production-ready agents with monitoring from day one.

Stack: Agno, BMasterAI

---

AI Real Estate Agent Team

August 2025

Multi-agent property search and analysis platform with comprehensive BMasterAI logging across agent performance, task execution, and market analysis flows.

Stack: Multi-agent, BMasterAI

---

Streamlit Business Consultant

July 2025

AI-powered business consultant with market analysis, competitor research, strategic recommendations, and risk assessment in a Streamlit UI.

Stack: Streamlit, BMasterAI

---

Google ADK Enterprise Consultant

July 2025

Enterprise-grade AI business consultant integrating Google's Agent Development Kit with BMasterAI monitoring and management.

Stack: Google ADK, BMasterAI

---

Minimal RAG

July 2025

Minimal working RAG implementation. Simplest possible starting point for retrieval-augmented agents.

---

Install

pip install bmasterai

Or from source:

git clone https://github.com/travis-burmaster/bmasterai.git
cd bmasterai
pip install -e .[dev]

Quickstart

from bmasterai.logging import configure_logging, get_logger, LogLevel, EventType
from bmasterai.monitoring import get_monitor

configure_logging(log_level=LogLevel.INFO, enable_console=True, enable_file=True)
logger = get_logger("my-agent")
monitor = get_monitor()

# Log agent events
logger.log_event(EventType.TASK_START, "Starting summarization task", {"model": "claude-3-5-sonnet"})

# Record metrics
monitor.record_metric("tokens_used", 1240)
monitor.record_metric("latency_ms", 850)

logger.log_event(EventType.TASK_COMPLETE, "Task done", {"success": True})

See examples/basic_usage.py for a full working example.

Documentation

Full API reference and deployment guides: README.content.md

Kubernetes deployment: README-k8s.md

Contributing

New examples welcome. Open a PR with:

• A clear learning objective
• Working code that runs end to end
• A README explaining the architecture and how to run it

License

MIT