The best agent frameworks, according to hackers
28 agent frameworks recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
- Most recommendedClaude Agent SDK36 people recommend it
- Built to lastLangChainowners report a median of 1 years
- Least regretHive agent framework10 recommend, 0 criticize
All agent frameworks
- Claude Agent SDKDevelopers praise it for building agents on top of Claude Code at a predictable cost. Critics agree the main caveat is Anthropic's shifting rules on using subscriptions with it.36recommend
- LangChainRecommended for easy switching between LLM providers and many integrations; critics find it bloated, overly abstract, and built for older models.32recommend
- MastraOpen-source TypeScript framework for building AI agents, praised for workflow primitives and developer experience. Common caveat: its opinionated patterns can slow you down when your project doesn't fit them.32recommend
- LangGraphRecommended for deterministic, stateful agent workflows with graph-based control, human-in-the-loop and time travel. Critics find it rigid and overly complex for simple tasks.31recommend
- DSPyPeople use it to structure LLM calls as typed programs and optimize prompts. Critics say its value depends on a good automated metric and up-front setup.29recommend
- Pydantic AIDevelopers like it for handling the tool-call loop, structured outputs and typed LLM calls with little to learn. Caveat: good observability requires the paid SaaS.27recommend
- OpenAI Agents SDKDevelopers like its abstractions for handoffs, tools and guardrails, and find it simple to start with. Common caveat: lock-in to OpenAI's platform and frequent API changes.22recommend
- Vercel AI SDKDevelopers use it to connect many AI model providers through one streaming interface, with well-written code. Critics question its single HTTP+SSE streaming transport design.19recommend
- Hive by AdenOpen-source framework for building autonomous AI agents, praised for its agent primitives and solid event-loop architecture.19recommend
- Google Agent Development KitFree open-source framework for building AI agents, praised for clean abstractions and multi-language support. Critics call it unpolished and rapidly changing.16recommend
- PipecatOpen-source framework for real-time voice agents, praised for flexibility and community; critics say deploying at scale needs significant work.16recommend
- Strands HarnessOpen-source agent framework praised for easy setup, pluggable design and not tying you to one model provider. Critics note it lacks durable execution and that its benchmark comparison is misleading.15recommend
- JidoFree Elixir framework for building AI agents, recommended for fitting the BEAM's fault tolerance and streamlining repeated agent workflows.15recommend
- Agency multi-agent platformDevelopers recommend this open-source multi-agent platform for autonomous coding, citing long-term use, fewer drift issues, and good results on parallel implementation tasks.14recommend
- BMAD MethodGives AI coding agents detailed rules, specs and agent roles for structured projects. Common caveat: it gets code most of the way to production-ready, and critics find its process heavy.14recommend
- Hive agent frameworkPeople recommend Hive for production-focused, self-improving AI agents with stateful execution and verification loops, especially for ERP-style work where accuracy matters.10recommend
- Google Agent Development Kit (ADK)Developers recommend it as a solid open-source framework for building agents across Python, TypeScript, Go and Java. Common caveat: Python dependency management is painful.8recommend
- Microsoft Agent FrameworkDevelopers recommend it for a solid orchestration layer and middleware that is easier to work with, comparable to LangChain and Pydantic-AI. Caveat: critics say it still just wraps LLM calls in more abstractions.8recommend
- Channels SDKLets any AI agent run in Slack, Teams and other chat apps, with quick setup and open-source use. Critics say the core service is closed and Team pricing deters some.8recommend
- RS-SDKPeople recommend it as a free, fun way to point AI agents at RuneScape, where agents have written their own scripts and finished quests. Critics note that botting breaks the game's terms of service.7recommend
- CrewAIOpen-source Python framework for building multi-agent AI workflows, cited as a common option next to LangGraph. Critics say it lacks durable execution and can feel heavy.6recommend
- Apache BurrFree framework for stateful AI agent workflows, with built-in observability; critics mostly object to its vibe-coded, AI-generated landing page.6recommend
- Claude Code SDKAnthropic's official SDK for building agentic coding tools; people recommend it for handling integration and agent use.6recommend
- OpenAI SDKDevelopers use it to call OpenAI models directly, without extra library layers. Critics note that calling the HTTP endpoint directly often makes the SDK unnecessary.5recommend
- smolagentsA lightweight Python framework for building AI agents, recommended as a simpler alternative to larger frameworks. Critics note that standard tool calling often beats its code-agent approach.5recommend
- HaystackOpen-source framework and editor that people recommend for agents, RAG and navigating complex codebases. The most common criticism targets deepset's branding rather than the software itself.5recommend
- MirascopeAn LLM toolkit that leaves the developer in control of the loop instead of imposing an agent loop.5recommend
- Output.aiOpen-source framework extracted from production AI agent work, praised for credential management that avoids plain .env files.5recommend