Skip to content

Agent SDKs

Agent SDKs are the engines behind AI coding agents. This section focuses on SDKs purpose-built for coding — frameworks designed to help LLMs read, write, debug, and ship code.

These SDKs are specifically built for software engineering agents — they understand codebases, file systems, git, and development workflows.

SDKByFocusPowers
AdalFlowSylph.AISelf-evolving agents, auto-optimizationAdaL CLI
OpenHands SDKOpenHandsCloud coding agent platformOpenHands (formerly OpenDevin)
SWE-agentPrinceton NLPAutonomous issue fixingSWE-bench SOTA
AiderPaul GauthierTerminal pair programmingAider CLI

Unlike general-purpose agent frameworks, coding agent SDKs provide:

  • Code-aware context — AST parsing, repo maps, dependency graphs
  • Edit formats — Structured diff/patch generation (search-replace, unified diff)
  • Sandbox execution — Safe code execution in containers
  • Git integration — Branch, commit, PR workflows built-in
  • SWE-bench tested — Validated on real GitHub issues

These are general agent frameworks that can be used to build coding agents, but aren’t coding-specific.

FrameworkByKey StrengthBest For
LangGraphLangChainGraph-based state machinesComplex multi-step workflows
OpenAI Agents SDKOpenAINative OpenAI integrationOpenAI-only agents
CrewAICrewAIRole-based multi-agent teamsTeam-based automation
Microsoft Agent FrameworkMicrosoftAutoGen + Semantic KernelEnterprise .NET
Pydantic AIPydanticType-safe, model-agnosticProduction APIs
SmolagentsHugging FaceLightweight, code-firstPrototyping
Strands AgentsAWSModel-driven, provider-agnosticMulti-agent systems
SDKByFocus
Semantic KernelMicrosoftEnterprise AI orchestration
LlamaIndex AgentsLlamaIndexRAG + agent workflows
AgnoAgnoLightweight multi-modal agents
MastraMastraTypeScript-first agents
Google ADKGoogleGemini-native agents
Building a coding agent?
├── Want self-evolving + auto-optimization? → AdalFlow
├── Need cloud sandbox execution? → OpenHands SDK
├── Fixing GitHub issues autonomously? → SWE-agent
├── Terminal pair programming? → Aider
├── Need general agent orchestration?
│ ├── Complex graph workflows → LangGraph
│ ├── OpenAI models only → OpenAI Agents SDK
│ ├── Multi-agent teams → CrewAI
│ └── Enterprise .NET → Microsoft Agent Framework
└── Lightweight / research → Smolagents or Pydantic AI