Kailash Kaizen — AI Agent Framework

Kaizen is a production-ready AI agent framework built on Kailash Core SDK. It provides signature-based programming, multi-agent coordination, autonomy infrastructure, and enterprise features (trust, governance, observability, cost tracking).

When to Use

Use Kaizen when asking about AI agents, agent framework, BaseAgent, multi-agent systems, agent coordination, signatures, RAG agents, vision agents, audio agents, multimodal agents, prompt optimization, chain of thought, ReAct pattern, Planning agent, PEV, Tree-of-Thoughts, pipeline patterns (supervisor-worker, router, ensemble, blackboard, parallel, sequential, handoff, consensus, debate), A2A protocol, streaming agents, agent testing, agent memory, agentic workflows, AgentRegistry, 100+ agents, capability discovery, fault tolerance, health monitoring, trust protocol, EATP, TrustedAgent, trust chains, secure messaging, credential rotation, cross-organization agents, agent manifest, TOML manifest, GovernanceManifest, deploy agent, FileRegistry, DAG validation, schema compatibility, cost estimation, composition validation, catalog server, budget tracking, BudgetTracker, posture budget, L3 autonomy, L3 primitives (EnvelopeTracker, EnvelopeSplitter, EnvelopeEnforcer, ScopedContext, ContextScope, ScopeProjection, DataClassification, MessageRouter, MessageChannel, DeadLetterStore, AgentFactory, AgentInstance, AgentInstanceRegistry, AgentSpec, PlanExecutor, PlanValidator, PlanModification, Plan DAG), gradient zone, agent spawning, cascade termination, L3Runtime, L3EventBus, L3EventType, EatpTranslator, governance events, GovernedSupervisor, governed agent, progressive disclosure, AccountabilityTracker, CascadeManager, ClearanceEnforcer, DerelictionDetector, BypassManager, VacancyManager, kaizen-agents, PACT governance, anti-self-modification, ReadOnlyView, NaN defense, bounded collections, Delegate, delegate facade, typed events (TextDelta, ToolCallStart, DelegateEvent), run_sync, multi-provider, StreamingChatAdapter, adapter registry (OpenAI, Anthropic, Google, Ollama), tool hydration, ToolHydrator, search_tools, BM25 search, incremental streaming, token streaming, wrapper composition (WrapperBase, L3GovernedAgent, MonitoredAgent, StreamingAgent, SupervisorWrapper, DuplicateWrapperError, WrapperOrderError, canonical stacking order, GovernanceRejectedError, BudgetExhaustedError, StreamEvent, run_stream), or provider protocols (StreamingProvider, ToolCallingProvider, StructuredOutputProvider, ProviderCapability, get_provider_for_model, LLMBased routing).

Quick Start

Basic Agent

from kaizen.core.base_agent import BaseAgent
from kaizen.signatures import Signature, InputField, OutputField
from dataclasses import dataclass

class SummarizeSignature(Signature):
    text: str = InputField(description="Text to summarize")
    summary: str = OutputField(description="Generated summary")

@dataclass
class SummaryConfig:
    llm_provider: str = os.environ.get("LLM_PROVIDER", "openai")
    model: str = os.environ["LLM_MODEL"]
    temperature: float = 0.7

class SummaryAgent(BaseAgent):
    def __init__(self, config: SummaryConfig):
        super().__init__(config=config, signature=SummarizeSignature())

agent = SummaryAgent(SummaryConfig())
result = agent.run(text="Long text here...")
print(result['summary'])

Pipeline Patterns (Orchestration)

from kaizen_agents.patterns.pipeline import Pipeline

# Ensemble: Multi-perspective collaboration
pipeline = Pipeline.ensemble(
    agents=[code_expert, data_expert, writing_expert, research_expert],
    synthesizer=synthesis_agent,
    discovery_mode="a2a",
    top_k=3,
)
result = pipeline.run(task="Analyze codebase", input="repo_path")

# Router: Intelligent LLM-based task delegation
router = Pipeline.router(
    agents=[code_agent, data_agent, writing_agent],
    routing_strategy="semantic",
)

See kaizen-key-concepts for the 9 pipeline patterns and the 6 autonomy subsystems.

Critical Rules

• Define signatures before implementing agents
• Extend BaseAgent for production agents
• Use type hints in signatures for validation
• Track costs in production environments
• Test agents with real infrastructure (Tier 2/3: NO mocking — rules/testing.md)
• Enable hooks for observability
• Use AgentRegistry for distributed coordination
• Use response_format for structured output (not provider_config)
• Set structured_output_mode="explicit" for new agents
• NEVER skip signature definitions
• NEVER ignore cost tracking in production
• NEVER put structured output keys in provider_config

Sub-File Index

Getting Started

LLM Wire Layer

Agent Patterns

Multi-Agent & Orchestration

Multimodal Processing

Advanced Features

Observability & Monitoring

Enterprise Trust & Governance

L3 Autonomy Primitives

Composition & Providers

Testing & Quality

Integration Patterns

With DataFlow

from kaizen.core.base_agent import BaseAgent
from dataflow import DataFlow

class DataAgent(BaseAgent):
    def __init__(self, config, db: DataFlow):
        self.db = db
        super().__init__(config=config, signature=MySignature())

With Nexus

from kaizen.core.base_agent import BaseAgent
from nexus import Nexus

agent_workflow = create_agent_workflow()
app = Nexus()
app.register("agent", agent_workflow.build())
app.start()  # Agents available via API/CLI/MCP

With Core SDK

from kailash.workflow.builder import WorkflowBuilder

workflow = WorkflowBuilder()
workflow.add_node("KaizenAgent", "agent1", {"agent": my_agent, "input": "..."})

Related Skills

• 01-core-sdk — Core workflow patterns
• 02-dataflow — Database integration
• 03-nexus — Multi-channel deployment
• 05-kailash-mcp — MCP server integration
• 17-gold-standards — Best practices

Support

• kaizen-specialist — Kaizen framework implementation
• testing-specialist — Agent testing strategies
• decide-framework skill — When to use Kaizen vs other frameworks