The same agent, on 10 frameworks
Every tutorial builds one research agent: it searches, keeps what several independent sources carry, and packs the result into a token budget. The framework is the only variable, so you can compare the frameworks rather than the examples.
Which one should you use?
Start from what the agent has to survive, not from the language.
- IfA run must survive a restart, or pause for a human, or be replayed
Durable state and checkpointing are the whole point of it. This is the 2026 production default for a reason.
- IfThe agent needs files and a shell as well as the web
Built-in Read, Write, Bash and Grep, plus MCP — so unlob arrives without writing a single tool wrapper.
- IfThe output feeds another system and must validate
A validated model rather than a string you parse and hope about.
- IfA user is waiting and should watch it work
Streaming and UI state are one system. Shortest path to something people can see.
- IfIt sits between other systems and runs on a schedule
And if the job is really "search hourly and post the new ones", consider dropping the model entirely — n8n will do that deterministically for no model cost.
Python
Pydantic AI
PydanticBuild a research agent on Pydantic AI with typed dependencies, validated structured output and unlob as a tool.
4 known gotchas · Python
CrewAI
CrewAIBuild a research crew on CrewAI, giving a researcher and a fact-checker separate roles over the same unlob index.
4 known gotchas · Python
LlamaIndex Workflows
LlamaIndexBuild a research agent as an event-driven LlamaIndex Workflow, with explicit steps for search, corroboration and synthesis.
4 known gotchas · Python
Google ADK
GoogleBuild a research agent on Google’s Agent Development Kit, wrapping unlob as a FunctionTool alongside Gemini models.
4 known gotchas · Python
TypeScript
Vercel AI SDK
VercelBuild a research agent with the Vercel AI SDK, defining unlob tools with Zod schemas and streaming the result to a React UI.
4 known gotchas · TypeScript
Mastra
MastraBuild a research agent on Mastra, combining typed tools, durable workflows and built-in memory in one TypeScript codebase.
4 known gotchas · TypeScript
Python and TypeScript
LangGraph
LangChainBuild a research agent on LangGraph, wiring unlob search and context assembly as tools on a prebuilt ReAct graph with checkpointed state.
4 known gotchas · Python and TypeScript
Claude Agent SDK
AnthropicBuild a research agent on the Claude Agent SDK, adding unlob as an MCP server alongside the SDK’s built-in file and shell tools.
4 known gotchas · Python and TypeScript
OpenAI Agents SDK
OpenAIBuild a research agent on the OpenAI Agents SDK, exposing unlob search and context assembly as function tools with handoffs and tracing.
4 known gotchas · Python and TypeScript
No code
Every tutorial runs on the free tier
10,000 requests a month, no card. Enough to build the agent, break it, and put a real evaluation set through it.