Skip to content

Features

Explore Agenkit's comprehensive feature set.

Overview

Agenkit provides production-ready features organized into five categories:

  • Transport Layer

    Connect agents across processes and languages with HTTP, gRPC, and WebSocket.

    Learn more

  • Middleware

    Add resilience, caching, rate limiting, and observability with composable middleware.

    Learn more

  • Composition

    Orchestrate multiple agents into workflows with sequential, parallel, and router patterns.

    Learn more

  • Observability

    Full OpenTelemetry integration with distributed tracing and Prometheus metrics.

    Learn more

  • LLM Adapters

    Connect to any LLM provider with a consistent interface. Swap providers with one line.

    Learn more

  • Tools

    Extend agents with executable tools for deterministic operations.

    Learn more

Feature Matrix

Feature Python Go Cross-Language
Transport Layer
HTTP/1.1 ✅ ✅ ✅
HTTP/2 ✅ ✅ ✅
HTTP/3 (QUIC) ✅ ✅ ✅
gRPC ✅ ✅ ✅
WebSocket ✅ ✅ ✅
Middleware
Retry ✅ ✅ N/A
Circuit Breaker ✅ ✅ N/A
Timeout ✅ ✅ N/A
Rate Limiter ✅ ✅ N/A
Caching (LRU) ✅ ✅ N/A
Batching ✅ ✅ N/A
Composition
Sequential ✅ ✅ ✅
Parallel ✅ ✅ ✅
Fallback ✅ ✅ ✅
Conditional ✅ ✅ ✅
Observability
OpenTelemetry Tracing ✅ ✅ ✅
W3C Trace Context ✅ ✅ ✅
Prometheus Metrics ✅ ✅ ✅
Structured Logging ✅ ✅ ✅
LLM Adapters
Anthropic (Claude) ✅ ⏳ N/A
OpenAI (GPT) ✅ ⏳ N/A
Google Gemini ✅ ⏳ N/A
AWS Bedrock ✅ ⏳ N/A
Ollama (Local) ✅ ⏳ N/A
LiteLLM (100+ providers) ✅ ⏳ N/A
Tools
Tool Interface ✅ ✅ N/A
Tool Registry ✅ ✅ N/A

Quick Examples

Transport Layer

# HTTP
agent = RemoteAgent(name="api", endpoint="http://localhost:8080")

# gRPC
agent = RemoteAgent(name="api", endpoint="grpc://localhost:50051")

# WebSocket
agent = RemoteAgent(name="api", endpoint="ws://localhost:8080")

Middleware

# Stack middleware
agent = MyAgent()
agent = CachingDecorator(agent, max_size=1000)
agent = RetryDecorator(agent, max_attempts=3)
agent = CircuitBreaker(agent)
agent = TimeoutDecorator(agent, timeout=30.0)

Composition

# Sequential pipeline
pipeline = SequentialAgent([validator, processor, formatter])

# Parallel analysis
analysis = ParallelAgent([sentiment, summary, entities])

# Router
router = ConditionalAgent(
    agents=[weather_agent, news_agent, general_agent],
    condition=route_function
)

Observability

# Initialize tracing and metrics
init_tracing("my-service", otlp_endpoint="http://jaeger:4317")
init_metrics("my-service", port=8001)

# Wrap agent
agent = TracingMiddleware(MetricsMiddleware(MyAgent()))

LLM Adapters

from agenkit.adapters.llm import AnthropicLLM, OpenAILLM, GeminiLLM

# Use Anthropic
llm = AnthropicLLM(api_key="sk-ant-...")

# Swap to OpenAI (same interface!)
llm = OpenAILLM(api_key="sk-...")

# Or try Gemini
llm = GeminiLLM(api_key="...")

# All use the same interface
response = await llm.complete(messages)

Performance

All features are designed for production use with minimal overhead:

  • Transport overhead: <1% in realistic workloads
  • Middleware overhead: <0.01% per middleware
  • Composition overhead: <5% for sequential/parallel
  • Observability overhead: <2% with tracing enabled

View detailed benchmarks →

Next Steps