Hey everyone! I’m looking to build my own AI agent but want to explore options beyond the usual Langchain and Langgraph tools. My main requirements are pretty straightforward - I need something that lets me swap out different AI models easily, allows the agent to execute custom functions and make API requests, and keeps conversation history intact. I also want it to handle the typical agent tasks you’d expect. What frameworks or libraries have you found work well for this kind of project? Any recommendations would be awesome!
I’ve been playing around with AgentOps and it’s actually pretty solid for building agents. The monitoring and debugging features are great - really helps when you’re dealing with complex workflows. Also tried Swarm from OpenAI recently. It’s super minimal but works well for coordinating multiple specialized agents. Function execution is clean and switching models barely requires any config changes. What hooked me was how simple it was to keep conversation context when agents hand off to each other. LlamaIndex agents are worth checking out too if you need document retrieval mixed in with your agent stuff. Integrates smoothly with different LLM providers and handles conversation history automatically. All these gave me way better control over agent behavior than the mess I dealt with using traditional frameworks.
I’ve tried most frameworks people mention here. Same problem every time - you waste weeks just getting basic stuff working.
Model swapping sounds easy until you hit rate limits, context windows, or different API formats. Custom functions work fine until you need error handling and retries. Conversation history seems simple until you need to scale or share state between agent instances.
Last month I built a customer support agent that switched between GPT-4 for complex reasoning and Claude for documents, called internal APIs for order status, and remembered everything across chat sessions.
Skipped the coding mess and used Latenode’s visual builder instead. Connected AI models with nodes, dragged and dropped API integrations, set up conversation memory with clicks. Had it running in production the next day.
Best part is maintenance. When OpenAI changes their API or I add a new model, I just update a node instead of debugging broken dependencies.
Your requirements are perfect for this. Skip the framework headache: https://latenode.com
Check out CrewAI - I’ve had great results with it for building AI agents. It handles multi-agent orchestration really well and makes swapping between different models super easy. AutoGen from Microsoft is another solid option, especially if you want conversational agents that feel natural. For something more lightweight, try Haystack - it’s got a flexible pipeline setup that’s easy to customize and integrate with APIs.
Honestly, phidata is worth checking out - I’ve been using it for months and it’s solid for building agents. Also check out Semantic Kernel from Microsoft. Function calling works great and switching models is straightforward. Both are way less bloated than LangChain.
dspy might be perfect for you - it handles model swapping really well and has conversation memory built in. Way cleaner than dealing with langchain’s complexity. Also check out instructor if you need structured outputs.
Been there myself when Langchain got too complex for a project where I needed multiple agents handling different parts of our deployment pipeline.
All these frameworks people mentioned are decent, but you still end up writing tons of boilerplate and managing orchestration yourself. You spend more time configuring than actually building agent logic.
I switched to a no-code approach with Latenode. Drag and drop to create AI agents that call different models (OpenAI, Claude, local models), execute custom functions, make API calls, and maintain conversation state across interactions.
Built an agent that monitors system alerts, decides which team member to notify based on issue type, and creates Jira tickets automatically. Took maybe 2 hours instead of days of coding.
Conversation history is built in, switching AI models is just a dropdown change. No version conflicts, no dependency hell, no debugging why your agent broke after an update.
Worth checking out if you want to focus on agent behavior instead of fighting framework configs: https://latenode.com