examples/

directory
v0.87.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jun 21, 2026 License: Apache-2.0

README

Agenkit Examples

Comprehensive collection of 285 examples across 6 languages (including WASM!) demonstrating all Agenkit features, patterns, and integrations.

🎯 Start Here

New to Agenkit? Follow this path:

  1. 01_basic_agent.py - Your first agent
  2. 02_sequential_pattern.py - Composing agents
  3. 03_parallel_pattern.py - Parallel execution
  4. 05_tool_usage.py - Adding tools
  5. 06_pattern_composition.py - Complex patterns

Want structured learning? Check out our Tutorials with Jupyter and Marimo notebooks!

📊 Quick Stats

Language Examples Status
Python 156 ✅ Complete
Rust 35 ✅ Complete (+ WASM!)
C++ 42 ✅ Complete
Zig 37 ✅ Complete
Go 7 ⚠️ Growing
TypeScript 8 ⚠️ Growing
Total 285

📂 Examples by Category

🧠 Core Patterns (11 examples)

The 11 foundational agent patterns from the Agenkit library.

Location: patterns/

Pattern Description Python Go TS
ReAct Reasoning + Acting loop ✅ ✅ ✅
Reflection Self-critique and improvement ✅ ✅ ✅
Agents-as-Tools Agents calling other agents ✅ ✅ ✅
Orchestration Coordinate multiple agents ✅ ✅ ✅
Conversational Stateful conversations ✅ ✅ ✅
Task Goal-oriented execution ✅ ✅ ✅
Multiagent Parallel agent collaboration ✅ ✅ ✅
Planning Multi-step planning ✅ - -
Autonomous Self-directed agents ✅ - -
Memory Hierarchy Tiered memory system ✅ - -
Reasoning with Tools Tool-augmented reasoning ✅ - -

Key Files:

  • patterns/react-pattern.py - Reason, Act, Observe loop
  • patterns/reflection-pattern.py - Self-improvement cycle
  • patterns/orchestration-pattern.py - Multi-agent coordination
  • patterns/multiagent-pattern.py - Parallel collaboration
🎨 Advanced Reasoning Techniques (6 examples)

Cutting-edge reasoning methods for complex problem-solving.

Location: techniques/reasoning/

Technique Description File
Chain-of-Thought (CoT) Step-by-step reasoning cot_example.py
Tree-of-Thought (ToT) Multi-path exploration tot_example.py
Self-Consistency (SC) Voting for reliability sc_example.py
Graph-of-Thought (GoT) Graph-based reasoning got_example.py
Least-to-Most (LTM) Decomposition strategy ltm_example.py
Plan-and-Solve Planning before execution plan_and_solve_example.py

Use Cases:

  • Complex problem decomposition
  • Multi-step reasoning
  • Uncertainty handling
  • Strategic planning

Learn More: Tutorial 03: Advanced Reasoning

🛡️ Production Middleware (8 examples)

Production-ready middleware for reliability, performance, and observability.

Location: middleware/

Middleware Purpose File
Retry Exponential backoff retry_example.py
Circuit Breaker Fail-fast pattern circuit_breaker_example.py
Timeout Request deadlines timeout_example.py
Caching LRU cache with TTL caching_example.py
Metrics Prometheus metrics metrics_example.py
Rate Limiter Token bucket rate_limiter_example.py
Batching Request batching batching_example.py
Per-User Rate Limiter User-specific limits per_user_rate_limiter_example.py

Key Features:

  • Stack multiple middleware
  • Zero-configuration defaults
  • Production-tested patterns
  • Full observability

Learn More: Tutorial 02: Production Patterns

🔧 Composition Patterns (4 examples)

Compose agents into complex workflows.

Location: composition/

  • Sequential: Pipeline processing (sequential_example.py)
  • Parallel: Concurrent execution (parallel_example.py)
  • Fallback: Graceful degradation (fallback_example.py)
  • Conditional: Dynamic routing (conditional_example.py)

Advanced Compositions: patterns/composition/

  • sequential-parallel.py - Hybrid workflows
  • router-supervisor.py - Routing + supervision
🌐 Transport Layer (2 examples)

Network protocols for distributed agents.

Location: transport/

  • gRPC: High-performance RPC (grpc_example.py)
  • WebSocket: Real-time bidirectional (websocket_example.py)

Cross-Language:

🔌 LLM Adapters (9 examples)

Connect to any LLM provider.

Locations:

Adapter Provider Files
OpenAI GPT-4, GPT-3.5 openai-basic.py, openai_example.py
Anthropic Claude 3.5 anthropic-basic.py, anthropic_example.py
Ollama Local models ollama-basic.py
LiteLLM 100+ providers litellm_providers.py
Remote HTTP API 01_basic_remote_agent.py

Advanced Features:

  • Streaming: streaming_example.py, 03_streaming.py
  • Provider Swapping: swapping_providers.py
  • Agent + LLM: agent_with_llm.py

Pattern Integration:

🧰 Tools (3 examples)

Extend agents with external capabilities.

Location: tools/

  • Database: SQL operations (database_example.py)
  • Search: Web search integration (search_example.py)
  • OS Tools: File system, shell commands (os_tools_example.py)

Example: 05_tool_usage.py

🔐 Safety & Permissions (4 examples)

Secure agents with validation and permissions.

Location: safety/

  1. Input Validation: Sanitize user input (01_input_validation.py)
  2. Output Validation: Filter agent responses (02_output_validation.py)
  3. Permissions: Fine-grained access control (03_permissions.py)
  4. Complete Stack: Production security setup (04_complete_safety_stack.py)

Features:

  • Content filtering
  • PII detection
  • Prompt injection defense
  • Rate limiting
  • Audit logging
🔒 Authentication (1 example)

Secure agent communications.

Location: auth/

  • Bearer Token: HTTP authentication (bearer_token_example.py)
🧭 Routing (2 examples)

Intelligent request routing.

Location: routing/

  • Semantic Tool Selection: Embedding-based routing (semantic_tool_selection_example.py)
  • Load Balancer: Distribute requests (load_balancer_example.py)
💾 Memory (1 example)

Stateful agents with memory.

Location: memory/

  • Conversational Agent: Context-aware conversations (conversational_agent.py)

See also: patterns/memory-hierarchy-pattern.py

💰 Budget & Cost Tracking (2 examples)

Monitor and control LLM costs.

Location: budget/

  • Cost Tracking: Monitor spending (cost_tracking_demo.py)
  • Extended Thinking: Token budget management (extended_thinking_demo.py)
💾 Checkpointing (1 example)

Durable agents with state persistence.

Location: checkpointing/

  • Durable Agent: Save/restore agent state (durable_agent_demo.py)
📊 Observability (1 example)

Monitor agents with OpenTelemetry.

Location: observability/

  • Full Stack: Tracing + Metrics + Logging (observability_example.py)

Features:

  • Distributed tracing
  • Prometheus metrics
  • Structured logging
  • Grafana dashboards

Learn More: Tutorial 04: Deployment

🧪 Evaluation & Testing (3 examples)

Test, optimize, and compare agents.

Location: evaluation/

  • Evaluation Demo: Quality metrics (evaluation_demo.py)
  • A/B Testing: Compare agent versions (ab_testing_demo.py)
  • Optimization: Hyperparameter tuning (optimization_demo.py)

Learn More: Tutorial 05: Testing Patterns

🎭 Pattern Usage Examples (7 examples)

Real-world pattern usage.

Location: patterns/usage/

  • sequential-usage.py - Sequential pipelines
  • parallel-usage.py - Parallel execution
  • fallback-usage.py - Error handling
  • router-usage.py - Dynamic routing
  • supervisor-usage.py - Supervised agents
  • collaborative-usage.py - Agent collaboration
  • human-in-loop-usage.py - Human feedback
🧪 Experimental Patterns (1 example)

Cutting-edge research implementations for validation and experimentation.

Location: experimental/

Recursive Language Models (RLM)

Directory: experimental/long_context_rlm/

Status: ⚠️ EXPERIMENTAL - Research validation, not production ready

Novel inference strategy from Zhang et al. (2025) that treats long contexts as external environment variables accessed programmatically via REPL.

Key Features:

  • Code Execution: Python REPL for context manipulation
  • Recursive Sub-calls: llm_query() function for delegating to sub-agents
  • Budget Protection: CostTracker + BudgetLimiter integration (high cost variance!)
  • Pattern Composition: Combines ReasoningWithTools, AgentTool, ReAct, BudgetLimiter

Paper Results:

  • BrowseComp+: 0% → 91% success rate
  • OOLONG: 0.04% → 58% success rate
  • Cost: Median comparable to baselines, 95th percentile 3-10x higher

Files:

  • basic_rlm.py - Core RecursiveREPLAgent implementation (364 LOC)
  • document_qa.py - Multi-document QA example (180 LOC)
  • test_with_api.py - Real API testing with budget protection
  • test_with_api_simple.py - Minimal test for debugging
  • API_TEST_RESULTS.md - Validation results with Anthropic Claude
  • ANALYSIS.md - Gap analysis and design decisions
  • tests/test_rlm_basic.py - 9 unit tests (all passing ✅)

Real API Test Results:

  • ✅ Code execution works
  • ✅ Recursive sub-calls work (llm_query())
  • ✅ Budget protection works
  • ⚠️ Prompt engineering needed (models not trained for RLM pattern)

Future Work: See Issue #473 for prompt engineering roadmap

Learn More:

🚀 Complete Applications (5 E2E examples)

Production-ready applications demonstrating full system architecture.

Location: e2e/

1. Code Review System

Directory: e2e/code-review/

Multi-agent code review with specialized reviewers.

Agents:

  • correctness_agent.py - Logic validation
  • performance_agent.py - Performance analysis
  • security_agent.py - Security scanning
  • style_agent.py - Style checking
  • synthesis_agent.py - Report generation

Orchestration: review_orchestrator.py

Total: 7 files

2. Customer Support System

Directory: e2e/customer-support/

RAG-based support with escalation.

Agents:

  • classifier.py - Intent classification
  • qa_agent.py - Q&A with knowledge base
  • escalation_agent.py - Human handoff
  • synthesis_agent.py - Response generation

Knowledge Base: vector_store.py

Total: 7 files

3. Research Assistant

Directory: e2e/research-assistant/

Multi-source research with memory.

Agents: research_agent.py Memory: memory_store.py (hierarchical memory) Tools: tool_registry.py (search, summarize, analyze)

Total: 6 files

4. LLM Optimizer

Directory: e2e/llm-optimizer/

Automatic prompt and model optimization.

Features:

  • Classification task optimization
  • Model selection (GPT-4, Claude, etc.)
  • Prompt tuning
  • A/B testing

Total: 3 files

5. Cross-Language System

Directory: e2e/cross-language-system/

Python orchestrator coordinating Go and TypeScript agents.

Architecture:

  • Python: orchestrator.py (coordination)
  • Go: gRPC services
  • TypeScript: WebSocket services

Total: 3 files (Python)

🌍 Multi-Language Examples
Rust Examples (35 files) ✨

Location: agenkit-rust/examples/

Full pattern coverage with evaluation framework:

  • All 11 core patterns (ReAct, Reflection, Orchestration, etc.)
  • 7 pattern usage examples
  • 8 evaluation examples (A/B testing, Bayesian optimization, etc.)
  • LLM adapters (OpenAI, Anthropic, Ollama, LiteLLM, Gemini, Bedrock)
  • HTTP transport
  • WASM support! wasm_browser_agent.html - Run agents in the browser

Why Rust?

  • Memory safety without garbage collection
  • Zero-cost abstractions
  • Excellent performance
  • Native WASM compilation for browser deployment
C++ Examples (42 files)

Location: agenkit-cpp/examples/

Full pattern coverage organized by category:

  • All 11 core patterns
  • Multiple adapters (OpenAI, Anthropic, etc.)
  • Evaluation framework
  • Integration examples
  • Production-ready error handling

Why C++?

  • Maximum performance for latency-sensitive applications
  • Systems programming capabilities
  • Existing C++ codebase integration
  • Fine-grained memory control
Zig Examples (37 files)

Location: agenkit-zig/examples/

Modern systems programming with full pattern coverage:

  • All 11 core patterns
  • Evaluation framework
  • Observability integration
  • LLM adapters
  • Integration examples

Why Zig?

  • Modern, simple syntax
  • C interoperability
  • Compile-time execution
  • Memory safety with manual control
Go Examples (7 files)

Location: go/

Reasoning techniques:

  • Chain-of-Thought (CoT)
  • Tree-of-Thought (ToT)
  • Self-Consistency (SC)
  • Graph-of-Thought (GoT)

Full implementation: agenkit-go/examples/

Why Go?

  • Simple concurrency with goroutines
  • Fast compilation
  • Excellent for microservices
  • Native gRPC support
TypeScript Examples (8 files)

Location: typescript/

Reasoning techniques + browser integrations:

Browser Examples: browser/

Full implementation: agenkit-ts/examples/

Why TypeScript?

  • Full-stack (Node.js + browser)
  • Type safety for JavaScript
  • Rich ecosystem
  • Web framework integration

🗺️ Learning Paths

Path 1: Beginner (1-2 hours)

Perfect for first-time users.

  1. Basic Agent → 01_basic_agent.py
  2. Sequential Pattern → 02_sequential_pattern.py
  3. Parallel Pattern → 03_parallel_pattern.py
  4. Tool Usage → 05_tool_usage.py
  5. LLM Integration → llm/openai_example.py

Next: Tutorial 01: Getting Started

Path 2: Production Engineer (2-3 hours)

For building production systems.

  1. Middleware Stack:

    • middleware/retry_example.py
    • middleware/circuit_breaker_example.py
    • middleware/caching_example.py
    • middleware/metrics_example.py

  2. Safety:

    • safety/04_complete_safety_stack.py

  3. Observability:

    • observability/observability_example.py

  4. E2E Application:

    • e2e/customer-support/ (complete system)

Next: Tutorial 02: Production Patterns

Path 3: AI Researcher (2-3 hours)

For advanced reasoning and experimentation.

  1. Core Patterns:

    • patterns/react-pattern.py
    • patterns/reflection-pattern.py
    • patterns/orchestration-pattern.py

  2. Reasoning Techniques:

    • techniques/reasoning/cot_example.py
    • techniques/reasoning/tot_example.py
    • techniques/reasoning/sc_example.py

  3. Evaluation:

    • evaluation/ab_testing_demo.py
    • evaluation/optimization_demo.py

Next: Tutorial 03: Advanced Reasoning

Path 4: Full-Stack Developer (3-4 hours)

For multi-language and distributed systems.

  1. Transport Layer:

    • transport/grpc_example.py
    • transport/websocket_example.py

  2. Adapters:

    • adapters/01_basic_remote_agent.py
    • adapters/03_streaming.py

  3. Cross-Language:

    • e2e/cross-language-system/

  4. Browser Integration:

    • browser/react-example/

Next: Tutorial 04: Deployment

🎯 Examples by Use Case

Building a Chatbot?
  1. patterns/conversational-pattern.py - Stateful conversations
  2. memory/conversational_agent.py - Memory management
  3. llm/streaming_example.py - Streaming responses
  4. e2e/customer-support/ - Complete support bot
Code Analysis?
  1. patterns/react-pattern.py - Reasoning about code
  2. patterns/reflection-pattern.py - Self-improvement
  3. e2e/code-review/ - Multi-agent code review
Research & Analysis?
  1. techniques/reasoning/cot_example.py - Chain-of-Thought
  2. techniques/reasoning/tot_example.py - Tree-of-Thought
  3. e2e/research-assistant/ - Full research system
Production Deployment?
  1. middleware/retry_example.py - Reliability
  2. middleware/circuit_breaker_example.py - Fail-fast
  3. observability/observability_example.py - Monitoring
  4. safety/04_complete_safety_stack.py - Security
  5. Tutorial 04: Deployment - Docker + K8s
Browser/WASM Deployment?
  1. Rust WASM Example - Run agents in browser
  2. browser/react-example/ - React integration
  3. browser/vue-example/ - Vue integration
  4. browser/svelte-example/ - Svelte integration

Why WASM?

  • Run AI agents directly in the browser (no server required)
  • Near-native performance
  • Privacy-preserving (data never leaves device)
  • Offline-capable applications
  • Cross-platform (desktop, mobile, web)

🚀 Quick Start

Run an Example
# Clone repository
git clone https://github.com/scttfrdmn/agenkit.git
cd agenkit/examples

# Install dependencies
pip install agenkit

# Set API keys (for LLM examples)
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."

# Run basic example
python 01_basic_agent.py

# Run pattern example
python patterns/react-pattern.py

# Run E2E application
cd e2e/customer-support
python main.py
Explore by Category
# Patterns
cd patterns && ls *.py

# Reasoning techniques
cd techniques/reasoning && ls

# Middleware
cd middleware && ls

# E2E applications
cd e2e && ls -d */
Documentation
Tutorials
Migration & Integration
More Examples

📊 Example Complexity Matrix

Category Beginner Intermediate Advanced Production
Patterns ✅ ReAct, Conversational ✅ Orchestration, Multiagent ✅ Planning, Autonomous -
Techniques - ✅ CoT, SC ✅ ToT, GoT, LTM -
Middleware ✅ Retry, Timeout ✅ Circuit Breaker, Caching ✅ Metrics, Rate Limiter ✅ Complete Stack
Transport - ✅ gRPC, WebSocket - ✅ Cross-language
E2E Apps - - ✅ Code Review, Support ✅ Research, Optimizer

🤝 Contributing

Found a bug or want to add an example?

  1. Report issues: GitHub Issues
  2. Discuss ideas: GitHub Discussions
  3. Contribute: CONTRIBUTING.md

📝 Example Template

Creating your own example? Use this structure:

"""
Brief description of what this example demonstrates.

Features:
- Feature 1
- Feature 2

Usage:
    python example_name.py
"""
from agenkit import Agent, Message

class YourAgent(Agent):
    @property
    def name(self) -> str:
        return "your-agent"

    async def process(self, message: Message) -> Message:
        # Your implementation
        return Message(role="assistant", content="response")

# Example usage
async def main():
    agent = YourAgent()
    response = await agent.process(Message(role="user", content="test"))
    print(response.content)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

🎉 Example Highlights

  1. patterns/react-pattern.py - ReAct loop
  2. llm/openai_example.py - OpenAI integration
  3. middleware/retry_example.py - Retry middleware
Most Complex
  1. e2e/code-review/ - 7-file multi-agent system
  2. e2e/customer-support/ - RAG + vectorstore + escalation
  3. e2e/research-assistant/ - Memory hierarchy + tools
Hidden Gems
  1. evaluation/optimization_demo.py - Hyperparameter tuning
  2. routing/semantic_tool_selection_example.py - Embedding routing
  3. budget/extended_thinking_demo.py - Token budget management

Ready to build? Start with 01_basic_agent.py or dive into E2E applications!

Made with ❤️ by the Agenkit community

Directories

Path Synopsis
apps
research-assistant/go/internal/agent
deployment
aws-lambda/go command
e2e
cross-language-system/go command
go
techniques/reasoning command
infrastructure
go command
Production-ready agent with load balancing, health checks, and enhanced retry.
 Production-ready agent with load balancing, health checks, and enhanced retry.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.