namzu.aidocs

Overview

Overview of the public surfaces exported by @namzu/sdk and how they fit together.

@namzu/sdk is the core Namzu package. It exposes the runtime primitives that stay stable across provider choices: agents, tools, stores, registries, sandboxes, RAG helpers, identifiers, and type contracts.

1. What Lives in the SDK

The SDK package is the place to start if you want to run agents without binding your application to a single model vendor.

CapabilityMain public surfaces
Agent executionReactiveAgent, PipelineAgent, RouterAgent, SupervisorAgent, AgentManager
Provider integrationProviderRegistry, LLMProvider, provider-facing type contracts
Tool systemdefineTool, ToolRegistry, getBuiltinTools, built-in filesystem and shell tools
Safety and isolationLocalSandboxProvider, verification and permission-related types
State and persistenceRunDiskStore, DiskTaskStore, conversation and memory stores
Session lifecycleInMemorySessionStore, DiskSessionStore, session hierarchy, handoff, retention
Prompt compositionSkillRegistry, assembleSystemPrompt, mergePersonas, withSessionContext
RetrievalTextChunker, DefaultRetriever, DefaultKnowledgeBase, createRAGTool
Desktop integration pointcreateComputerUseTool, ComputerUseHost types
Connectors and MCPConnectorRegistry, ConnectorManager, MCPClient, MCPToolDiscovery, MCPConnectorBridge
Wire bridgesmapRunToStreamEvent, A2A helpers (telemetry ships from @namzu/telemetry as of 0.4.0)

2. What Does Not Live in the SDK

The SDK intentionally leaves vendor-specific integrations and optional capability packages outside core:

  • Provider implementations live in their own packages under @namzu/*.
  • Desktop control lives in @namzu/computer-use.
  • Local-only workspace packages that are not published are not part of this docs surface.

This separation keeps the SDK install surface smaller and lets you compose only the pieces you actually deploy.

3. Core Concepts

The public package surface is easiest to understand through four concepts:

ConceptRole
ProviderSupplies chat() and chatStream() through the LLMProvider contract
AgentOwns the runtime behavior for a run and produces a result from messages plus config
Tool registryHolds callable tools and controls their availability state
Run contextCarries model settings, IDs, working directory, and optional runtime overrides

Three IDs are operationally important in 0.2.x runtime usage:

  • projectId identifies the long-lived project scope.
  • sessionId identifies the immediate run session.
  • tenantId identifies the isolation boundary.

4. SDK Docs Structure

The SDK docs are grouped by domain instead of a flat file list:

FolderPurpose
agents/agent classes, orchestration, delegation, and manager-facing behavior
runtime/runtime overview, IDs, configuration, and low-level query entrypoints
tools/tool definitions, built-ins, and safety policy
provider-integration/SDK-level provider registry and direct provider operations
integrations/connectors, MCP, plugins, and event bridges
prompting/skills, personas, and prompt composition
retrieval/knowledge bases, retrieval, and RAG surfaces
sessions/session hierarchy, workspaces, summaries, and retention
observability/telemetry and instrumentation
architecture/deeper source-tree and pattern-level architecture docs

5. Recommended Entry Pages

Read these pages next depending on what you are doing:

If you want to...Read
Run a first agentSDK Quickstart
Understand runtime componentsSDK Runtime
Understand provider registration and creationSDK Provider Integration
Choose the right agent class and delegation boundarySDK Agents
Build system prompts from persona and skill filesSDK Prompting
Add knowledge-base retrievalSDK Retrieval
Persist project-session-sub-session stateSDK Sessions
Define custom toolsSDK Tools
Integrate connectors, plugins, or MCP serversSDK Integrations
Configure tracing and metricsSDK Observability
Understand folder boundaries and architecture patternsSDK Architecture
Pick a model integrationProviders Overview
Add desktop controlComputer Use

On this page