Langchain/Langsmith Agentic Implentation #149
Replies: 1 comment 3 replies
|
Interesting idea! Our agents are tool-agnostic system prompts — they work with any LLM backend. A LangChain/LangSmith integration would be more of a runtime wrapper than a conversion target. If you have a specific format in mind (e.g., LangChain agent config files), feel free to open a PR adding a converter to |
|
no its much to complicated for that, but i did create a zip bundle for you, if you can translate that to your current system more power to ya. this zip is meant to be extracted into your repo, you point your agent at the start here.md, the easiest implementation is straight github, let the sub agents be your fleet of ai. this setup will take you from zero to a full highest tech stack implementation of lang chain/lang smith, this is something is usually found at the corporate level and many corporations aren't rocking this setup. this gives your agent hard core persistant memory, it will allow you to setup a pipeline agentic fleet and implement learning datasets (via pytorch) so your agents get smarter and can accomplish what ever goal your pipeline serves. The reason im handing this out is i really like the volume and complexity of your agents i will definitely use them in my fleet. Also I like the idea of putting corporate level power in the hands of individuals. do what you will with this just give me credit if you publish it thats all i ask. |
|
actually that kit is not as instructional as it should be i revamped it, here this one is complete |
|
as an addendum here is a zip with full instructions to clean your repo of hallucinations, this comes in handy. hallucination-clean-full-memory-birth-kit.zip |
Uh oh!
There was an error while loading. Please reload this page.
Instead of using .md files as the profiles for your agents, have you condidered using Langsmith/langchain implementation? I have implemented the the highest level of this technology stack and no joke i am able to do the work of 100 phd level programmers with my fleet of ai's all connected with agentcomms, persistent memory and the ability to learn. You would just have to have your ai write the scripts to install the prompts that give each ai its identity and a few other steps and you can do amazing things. your agents can hand off tasks to one another with agentcomms. i do see you have a make shift version of this, but checkpointing with postgrsSQL and lancedb for persistant memory and of course langchain/langsmith implementation would make your system bulletproof and like i said they would become learning ai, with semantic memory. i use github copilot and with the new hooks this works perfectly. claude already has hooks so that would work but claude is super buggy. right now the OP is GPT 5.4, it kills the competition. thats why a github copilot is a good solution you can switch agents on a dime. course i use cheap google subagents for my fleet but they are sufficiently guard railed, they do quite well, and when you run a huge fleet you cannot run top tier models unless you want to pay bank. Id be happy to show you my implementation sometime if you like, i have an Agent Almanac im working on that you could just point your ai at and it will set everthing up as long as you have the keys in .env. If you used that and loaded it with your agent profiles your system would be super robust. i have a bunch of stuff in my repos but it would be best to see my Agent Almanac when its done. but i suggest taking a look at langchain/langsmith that is where its at right now. as far as i know no individual has a setup like mine only corporations. but its not expensive, it takes time to understand and perfect that is certain. part of the technology is running traces, it enables you to take your fleet back in time, alter your code and then run forward and achieve a different result, its quite spectacular! I have pipelines where my agents are doing spectral analysis on 3d parts, this would take ages without them. anyhow i hope you try this it will make your system next level. i was thinking of offering this to people mysefl to level the playing field.
All reactions