Skip to content
Projects
/harness
Strands harness is a state-of-the-art, fully assembled agent harness.
/harness-sdk
The Strands Harness SDK for building an agent harness from the ground up.
/box
An agent sandbox that pairs OS containment with Dogwood policy.
/shell
A virtual shell designed for AI agents to use safely.
/evals
Validate your agent before you ship.
/labs
Experimental projects pushing the boundaries of what agents can do.
Learn
Courses
Guided lessons from your first agent to multi-agent systems.
Examples
Runnable samples you can drop into your own project.
Integrations
Contribute
Blog
Search
TS
TypeScript
Python
14,800+
TS
TypeScript
Python
Projects
/harness
Strands harness is a state-of-the-art, fully assembled agent harness.
Get started
Overview
Quickstart
Compose with the Strands Harness SDK
Configure the agent
Choose a model and reasoning
Add tools and instructions
Delegate to subagents
Load agent skills
Persist sessions
Give long-term memory
Checkpoints vs memory
Manage context and caching
Gate tool calls with interventions
Connect MCP servers
Run work in the background
Built-in tools
Shell and file tools
Web access
Programmatic tool calling
Task tracking and environment
Take Strands harness to production
Reference
Configuration reference
Versioning & Support
/harness-sdk
The Strands Harness SDK for building an agent harness from the ground up.
Get started
Overview
Quickstart
Quickstart
AI coding agents
Migrate
Choosing an agent foundation
Migrate from LangGraph
Migrate from OpenAI
Build guides
Tools
Overview
Attach and invoke tools
Use MCP tools
Create custom tools
Sessions
Persist state across sessions
Memory
Overview
Control what the agent remembers
Responses
Return structured output
Stream responses
Manage the context window
Pause for input and control
Coordinate multiple agents
Run guides
Deploy to production
Operating Agents in Production
Amazon Bedrock AgentCore
Overview
Python
TypeScript
AWS Lambda
AWS Fargate
AWS App Runner
Amazon EKS
Amazon EC2
Docker
Overview
Python
TypeScript
Kubernetes
Terraform
Nx Plugin for AWS
Observe your agent
Observability
Metrics
Traces
Logs
Secure for production
Responsible AI
Guardrails
Prompt Engineering
Trusted Message History
PII Redaction
Components
Agent loop
Production lifecycle controls
State
Storage
Overview
Search Strategies
Snapshots
Prompts
Hooks
Hook events
Context management
Overview
Built-in Modes
Custom Strategies
Strategy Presets
Context Estimation
Retry strategies
Interrupts
Overview
Interrupts in Multi-Agent Systems
Models
Overview
Model Routing
Amazon Bedrock
Bedrock Prompt Caching
Amazon Nova
Anthropic
Google
LiteLLM
llama.cpp
LlamaAPI
MistralAI
Ollama
OpenAI
OpenAI Responses API
SageMaker
Vercel
Writer
Custom Providers
Tools
Tool result format
MCP Transports
Executors
Community Tools Package
Vended Tools
Memory
Test Store
Bedrock Knowledge Base
Plugins
Overview
Skills
Context Offloader
Context Injector
GoalLoop
Build a custom plugin
Interventions
Overview
Human in the loop
Cedar Authorization
Steering
Sandbox
Overview
Available Sandboxes
Building a Custom Sandbox
Multi-agent
Agent2Agent (A2A)
A2A server configuration
Agents as Tools
Swarm
Graph
Graph components
Workflow
Streaming
Async Iterators
Callback Handlers
Event types
Voice
Overview
Quickstart
BidiAgent
Models
Overview
Bedrock Nova Sonic
Google Gemini Live
OpenAI Realtime
Tools
I/O
Input Content
Stream Events
Hooks
Session Management
Interrupts
Observability
Conversation management
Experimental
AgentConfig
Reference
Versioning & Support
API reference
API reference
/box
An agent sandbox that pairs OS containment with Dogwood policy.
Get started
Overview
Getting started
How-to guides
Run the Strands harness
Run Claude Code
Run Codex CLI
Run pi
Grant files and directories
Allow network destinations
Add a tool
Write a policy
Write shell policy
Record telemetry
Collect CLI traces and metrics
Security
Security model
Agent and tool sandboxes
Reference
box.toml
CLI
Policy actions
Strands Shell
Python (Monty)
/shell
A virtual shell designed for AI agents to use safely.
Get started
Overview
Quickstart
How it works
Guides
Configure the sandbox
Run the MCP server
Inspect the configuration
Components
Commands
Shell language and builtins
Lua scripting
Security
Security model
Reference
Reference
/evals
Validate your agent before you ship.
Get started
Overview
Quickstart
How evaluation works
Evaluate with AI
Build guides
Generate test cases
Plan topic coverage
Run guides
Task decorator
Result caching
Experiment management
Serialization
Evaluating remote traces
AgentCore evaluations
Components
Evaluators
Overview
Quality
Output
Trajectory
Interactions
Helpfulness
Faithfulness
Correctness
Coherence
Conciseness
Response relevance
Safety
Harmfulness
Refusal
Stereotyping
Multimodal
Multimodal output
Multimodal overall quality
Multimodal correctness
Multimodal faithfulness
Multimodal instruction following
Agentic
Instruction following
Goal success rate
Failure communication
Partial completion
Recovery strategy
Tool selection accuracy
Tool parameter accuracy
Skill
Skill selection accuracy
Skill instruction following
Deterministic
Custom
Detectors
Overview
Failure detection
Root cause analysis
Session diagnosis
Simulators
Overview
User simulation
Customizing user simulation
Tool simulation
Red teaming
Overview
Quickstart
Attack strategies
Writing custom cases
Scoring attacks
Reading the report
Chaos testing
CLI
Overview
run
generate
diagnose & fetch
report & validate
/labs
Experimental projects pushing the boundaries of what agents can do.
Learn
Courses
Guided lessons from your first agent to multi-agent systems.
Examples
Runnable samples you can drop into your own project.
Integrations
Contribute
Blog
TS
TypeScript
Python
Search
Theme
GitHub
Discord
Search
Ctrl
K
Cancel
/harness-sdk
Overview
Classes
AfterInvocationEvent
AfterModelCallEvent
AfterToolCallEvent
AfterToolsEvent
Agent
AgentMetrics
AgentResult
AgentResultEvent
AgentTrace
AudioBlock
BedrockModel
BeforeInvocationEvent
BeforeModelCallEvent
BeforeToolCallEvent
BeforeToolsEvent
CachePointBlock
CitationsBlock
ClassifierStrategy
ConcurrentInvocationError
ConcurrentToolExecutor
ConstantBackoff
ContentBlockEvent
ContextWindowOverflowError
ConversationManager
DecisionModel
DefaultModelRetryStrategy
DefaultNotConfiguredError
DocumentBlock
ExponentialBackoff
ExtractionTrigger
FallbackStrategy
FileStorage
FunctionTool
Graph
GuardContentBlock
HookableEvent
HookRegistry
ImageBlock
InitializedEvent
InterruptError
InterruptEvent
InterruptResponseContent
IntervalTrigger
InterventionHandler
InvocationTrigger
JsonBlock
JsonValidationError
LinearBackoff
LLMDecisionModel
MaxTokensError
McpClient
McpTaskCancelledError
McpTaskFailedError
MemoryManager
Message
MessageAddedEvent
Model
ModelContentBlockDeltaEvent
ModelContentBlockStartEvent
ModelContentBlockStopEvent
ModelError
ModelExtractor
ModelMessageEvent
ModelMessageStartEvent
ModelMessageStopEvent
ModelMetadataEvent
ModelRedactionEvent
ModelRetryStrategy
ModelRouter
ModelStreamUpdateEvent
ModelThrottledError
NullConversationManager
PosixShellSandbox
ReasoningBlock
RoutingCandidate
S3Location
Sandbox
SandboxAbortError
SandboxPathNotFoundError
SandboxTimeoutError
SequentialToolExecutor
SessionManager
SlidingWindowConversationManager
StateStore
StorageError
StreamEvent
StructuredOutputError
SummarizingConversationManager
Swarm
TextBlock
Tool
ToolNotFoundError
ToolResultBlock
ToolResultEvent
ToolStreamEvent
ToolStreamUpdateEvent
ToolUseBlock
ToolValidationError
Uncertain
VideoBlock
ZodTool
Interfaces
AddMessagesContext
AgentAsToolOptions
AgentMetadata
AudioBlockData
BackgroundTasksConfig
BackoffContext
BackoffStrategy
BaseModelConfig
BedrockGuardrailConfig
BedrockGuardrailRedactionConfig
BedrockModelConfig
BedrockModelOptions
CacheConfig
CachePointBlockData
Citation
CitationsBlockData
CitationsDelta
ClassifierStrategyOptions
ConstantBackoffOptions
ContextManagerConfig
ContextState
ContextStrategy
CountTokensOptions
DecisionResult
DefaultModelRetryStrategyOptions
DirectToolCallOptions
DocumentBlockData
ExecuteOptions
ExecuteToolContext
ExecuteToolResult
ExecutionResult
ExponentialBackoffOptions
ExtractionConfig
ExtractionResult
ExtractionTriggerContext
Extractor
ExtractorContext
FileInfo
FunctionToolConfig
GuardContentBlockData
GuardContentImage
GuardContentText
HookCallbackOptions
ImageBlockData
InjectionConfig
InjectionContext
Interrupt
InterruptParams
InterruptResponse
InterruptResponseContentData
IntervalTriggerOptions
InvokableTool
InvokeModelContext
InvokeModelResult
InvokeOptions
LifecycleObserver
LinearBackoffOptions
LLMDecisionModelOptions
LocationData
Logger
McpCallToolOptions
McpCancelledTask
McpCancelTaskResult
McpClientCredentials
McpClientOptions
McpCompletedTask
McpFailedTask
McpInputRequests
McpInputRequiredTask
McpInputResponses
McpListOptions
McpListToolsOptions
McpLoadServersOptions
McpProgress
McpServerConfig
McpTask
McpTaskError
McpTaskRequestOptions
McpTaskResult
McpToolFilters
McpUpdateTaskResult
McpWorkingTask
MemoryAddOptions
MemoryAddToolConfig
MemoryEntry
MemoryInjectionConfig
MemoryManagerConfig
MemoryMessageFilter
MemorySearchOptions
MemoryStore
MemoryStoreConfig
MemoryToolConfig
MessageData
Metrics
MiddlewareInterruptible
MiddlewareInterruptResult
MiddlewareStage
ModelContentBlockDeltaEventData
ModelContentBlockStartEventData
ModelExtractorOptions
ModelMessageStartEventData
ModelMessageStopEventData
ModelMetadataEventData
ModelRedactionEventData
ModelRouterOptions
ModelStopResponse
OffloadConditions
OffloadStrategyBuilder
OutputFile
Plugin
Question
ReasoningBlockData
ReasoningContentDelta
RedactInputContent
Redaction
RedactOutputContent
RoutingAttempt
RoutingCandidateOptions
RoutingContext
RoutingStrategy
S3LocationData
SearchOptions
SerializableMcpToolFilters
SessionManagerConfig
Snapshot
SnapshotManifest
SnapshotStorage
SnapshotTriggerParams
StashConfig
Storage
StreamChunk
StreamOptions
SummarizeConfig
TasksConfig
TextBlockData
TextDelta
ToolContext
ToolHandle
ToolResultBlockData
ToolSpec
ToolStreamEventData
ToolUse
ToolUseBlockData
ToolUseData
ToolUseInputDelta
ToolUseStart
TruncateConfig
Usage
VideoBlockData
ZodToolConfig
Type Aliases
AgentConfig
AgentStreamEvent
AnswerOf
AudioFormat
AudioSource
AudioSourceData
BedrockCacheConfig
BedrockCacheTTL
CandidateInput
CitationGeneratedContent
CitationLocation
CitationSourceContent
ContentBlock
ContentBlockData
ContentBlockDelta
ContentBlockStart
ContextManagerStrategy
ConversationManagerOptions
ConversationManagerReduceOptions
DecisionInput
DocumentContentBlock
DocumentContentBlockData
DocumentFormat
DocumentSource
DocumentSourceData
ElicitationCallback
ElicitationContext
FunctionToolCallback
GuardImageFormat
GuardImageSource
GuardQualifier
HookableEventConstructor
HookCallback
ImageFormat
ImageSource
ImageSourceData
InjectionTrigger
InterruptSource
InvocationState
InvokeArgs
JitterKind
JSONSchema
JSONValue
McpClientConfig
McpConnectionState
McpCreateTaskResult
McpDetailedTask
McpDirectCallToolResult
McpGetTaskResult
McpInputRequest
McpInputResponse
McpProgressCallback
McpSubmitToolResult
McpTaskStatus
McpToolFilterCallback
McpToolMatcher
McpTransport
MemoryContentBlockType
MiddlewareHandler
MiddlewareHandlerOf
MiddlewareInputHandler
MiddlewareNext
MiddlewareNextOf
MiddlewareOutputHandler
ModelStreamEvent
MultiAgentSaveLatestStrategy
OffloadTarget
OnError
ProactiveCompressionConfig
RetryDecision
RetryStrategy
Role
SaveLatestStrategy
Scope
SessionStorage
SlidingWindowConversationManagerConfig
SnapshotField
SnapshotLocation
SnapshotPreset
SnapshotTriggerCallback
StopReason
StrategyPresetName
StreamType
SummarizingConversationManagerConfig
SystemContentBlock
SystemPrompt
SystemPromptData
TakeSnapshotOptions
ToolCallerProxy
ToolChoice
ToolExecutorStrategy
ToolList
ToolResultContent
ToolResultStatus
ToolStreamGenerator
VideoFormat
VideoSource
VideoSourceData
Functions
configureLogging
contentBlockFromData
defineQuestions
isModelStreamEvent
tool
toolResultContentFromData
Select theme
Dark
Light
Auto
ImageFormat
type
ImageFormat
=
typeof
IMAGE_FORMATS
[number];
Defined in:
src/mime.ts:26