asaddhamani

I’ve been building a shared memory layer across all AI tools

www.memoryplugin.com

They also posted on Instagram saying autonomous killing would hurt Americans. So non American people don’t matter?

Or economic ruin, societal ruin they may cause along the way, like OpenAI.

yes thats the one

The iOS compass redesign is particularly egregious

Call me cynical but I think designers need to occasionally break things that were already solved long ago to justify their continued relevance. Explains a lot of redesigns that make things only worse, reshuffling interfaces, hiding things behind menus in form over function redesigns, etc.

MemoryPlugin (https://www.memoryplugin.com)

Long term memory for dozens of AI tools, designed for power users who want more control and flexibility than native memory systems and who do not want to be locked into any one platform. You can also have the system remember your entire chat history going back years and use this information to help you better in new chats, it sometimes makes chats 10x more useful when I say something like: “Using recall tool, do 10+ calls for 1000 tokens context each to learn about my interests, strengths, curiosities, what I’ve tried in the past, what worked, what didn’t, etc and suggest a new hobby I would enjoy”.

Without long term recall, AI is a super intelligence in your hands that uses the knowledge of the world to give you generic, nearly useless advice because of how generic it is. With long term memory, you have a super-intelligence that knows YOU. This is what MemoryPlugin solves for.

I love Tower and have paid for it for years. I can’t imagine using the git CLI now. GUIs were invented for a reason and the git CLI has terrible ergonomics and many ways to make costly mistakes.

I’ve been working on MemoryPlugin (https://www.memoryplugin.com), a tool that adds long term memory across AI tools

Lately I’ve worked on a chat history based memory feature that can recall information from every conversation you’ve ever had with ChatGPT and Claude. It’s been particularly useful and also technically fun to implement. Speed has been very important as I do just in time summarisation and a multi stage RAG pipeline, and most LLMs have unacceptable performance. I ended up going with GPT-OSS on Groq due to its ultra low latency often completing full generations before Gemini or ChatGPT APIs return even the first token.

The ability to recall details from conversations going back years makes tasks where I want personalised plans or feedback like 10x more useful, at times I get the AI to ingest tens of thousands of tokens of context to help me better.

I find the python tooling so confusing now. There’s pip, virtualenv, pipx, uv, probably half a dozen others I’m missing. I like node, npm isolates by default, npx is easy to understand, and the ecosystem is much less fragmented. I see a python app on GitHub and they’re all listing different package management tools. Reminds me of that competing standards xkcd.

I've been working on a tool to add long term memory to various AI tools. It started last year as a small scratch-your-own-itch side project. After using ChatGPT Plus for a month, I went back to TypingMind, my go-to AI client at the time, but I really missed the memory feature and wanted it there. So I made a simple memory plugin for it.

Over time, the project has grown to now support more than 17 platforms, thousands of users, and has been growing organically.

As of most recently a major feature I've been working on is full chat history based memory. Being able to remember and recall every conversation you've ever had across multiple supported AI tools, similar to the reference past conversations feature in various AI apps. This has been pretty intense and fun. Ingesting tens of millions of tokens per user, and doing complex multi-stage RAG on-the-fly across this vast dataset with a tight latency target for UX.

The project is MemoryPlugin: https://www.memoryplugin.com

Another project is a RAG app that's built specifically for books. No "we work with your receipts, and legal documents, instruction manuals, product documentation, lecture transcripts, your dogs novel, the script for a play and everything else possible". I wanted something tailored for books, specifically, non-fiction books. When you try to work everywhere, you can not deliver an amazing experience for any one specific use case. AskLibrary is tailored for non-fiction books, so everything from the answer generation pipelines, to the ingestion pipeline, and various other features are all designed for this specific use case. https://www.asklibrary.ai

I think a few of the things you’ve mentioned in the ChatGPT article are hallucinations. There’s no user interaction metadata about topics, average message length etc., you asked the AI and it gave you a plausible sounding answer. Also the memories / snippets of past conversations aren’t based on like last 30 conversations or so and they aren’t provided to every message. They are doing some kind of RAG prompt injection and they remove the injected context in the next message to not flood the context window. The AI itself seems to have no control over what’s injected and when, it’s a separate subsystem doing that injection.

In my experience, Qwen 3 coder has been very good for agentic coding with Cline. I tried DeepSeek v3.1 and wasn't pleased with it.

Claude now does have a weekly limit so if you are able to hit your weekly (undisclosed, dynamic) limit in 2 days, you're unable to use the services for the next 5 days. That is what Cerebras is referencing with no weekly limits. Claude has session count limits, dynamic limits within each session, and now weekly limits on top of all that.

I find it interesting, how OpenAI came out with a $200 plan, Anthropic did $100 and $200, then Gemini ups it to $250, and now Grok is at $300.

OpenAI is the only one that says "practically unlimited" and I have never hit any limit on my ChatGPT Pro plan. I hit limits on Claude Max (both plans) several times.

Why are these companies not upfront about what the limits are?

I can't trust MTEB as there's been a huge difference between benchmark scores and actual performance.

I made a small tool to help me compare various embedding models: https://www.vectorsimilaritytest.com/

Qwen embedding models score very highly but are highly sensitive to word order (they use last token pooling which simplified means they only look at the last word of the input). Change the word order and the scores change completely. Voyage models score highly too, but changing a word from singular to plural can again completely change the scores.

I find myself doing a hybrid search, rerank and shortlist the results, then feed them to an LLM to judge what is and isn't relevant.

API prices are way higher than actual inference cost.

DeepInfra is amazing in terms of price, like really, they have the Qwen3 embedding model for $0.002 per mn tokens. That's an order of magnitude cheaper than most alternatives with better benchmark scores. But the performance P99 is slow and the variance is huge. For latency sensitive workloads it's problematic, if they can fix that it'll be a no-brainer to use them. DeepInfra does tend to have the lowest prices of any API provider.

The pricing is a little on the higher side. Working on a performance-sensitive application, I tried Mercury and Groq (Llama 3.1 8b, Llama 4 Scout) and the performance was neck-and-neck but the pricing was way better for Groq.

But I'll be following diffusion models closely, and I hope we get some good open source ones soon. Excited about their potential.

In my experience, there are areas where Gemini did well but Claude didn't, same for o1 pro or o3, but for 90% of the work, I find Claude way more trustworthy, better at following instructions, not making syntax mistakes, etc. Gemini 2.5 Pro is way better than all their prior models, but I don't get the hype about it being a coding superstar. It's not bad, but Sonnet is still the primary workhorse. Sonnet is more expensive, so if Gemini was at the same level I'd be happy to save the money, but unfortunately, I've tried it with various approaches, played with the temperature, but in the vast majority of cases Claude does a better job.

Nope, it can be even a dozen (because agentic). Claude usage limits are actually based on token usage, and Claude Code uses a mix of Haiku and Sonnet. So your limits are split among those two models. I gave an estimation of how much usage you can expect in another comment on this thread, but you will find it hard to max out the $100 plan unless you are using it very, very extensively.

I wrote about this on my blog: https://www.asad.pw/llm-subscriptions-vs-apis-value-for-mone...

But basically you get ~300Mn input tokens and ~100Mn output tokens per month with Sonnet on the $100 plan. These are split across 50 sessions you are allowed, each session is 5 hrs starting from the first time you send a message until 5 hrs after the first message. During this time, you get ~6Mn input and ~2Mn output tokens for Sonnet. Claude Code seems to use a mix of Sonnet and Haiku, and Haiku has 2x the limits of Sonnet.

So if you absolutely maxed out your 50 sessions every month, that's $2400 worth of usage if you instead had used the API. So it's a great deal. It's not $100 worth of API credits you're buying, so they don't run out like that. You can exhaust limits for a given session, which is at most a 5 hr wait for your next one, or you can run out of 50 sessions, I don't know how strongly they enforce that limit and I think that limit is BS, but all in all the value for money is great, way better than using the API.

Sonnet in Copilot is crippled, Copilot agent mode is also very basic and failed every time I tried it. It would have been amazing 2 years ago, but now it's very meh.

GitHub is losing money on the subs, but they are definitely trying to reduce the bleed. One way to do that is to cut corners with LLM usage, by not sending as much context, trimming the context window, capping output token limits, these are all things Cursor also does btw, hence why Cline, with almost the same tech (in some ways its even inferior tech) achieves better results. I have hit $20 in API usage within a single day with Cline, Cursor lets you have "unlimited" usage for $20 for a month. So its optimised for saving costs, not for giving you the best experience. At $10 per month for Copilot, they need to save costs even more. So you get a bad experience, you think its the AI that is not capable, but the problem is with the companies burning VC money to corner the market, setting unrealistic expectations on pricing, etc.

I don’t know why people expect unlimited usage for limited cost. Copilot hasn’t been good for a long time. They had the first mover advantage but they were too slow to improve the product. It’s still not caught up to cursor or windsurf. Cline leaves it so far in the dust it’s like a decade behind in AI years. So you get what you pay for.

Claude is still the gold standard for AI assisted coding. All your Geminis and o3s of the world still don’t match up to Claude.

I started using Claude code once it became a fixed price with my Claude max subscription. And it’s taken a little getting used to vs Cline, but I think it’s closer to Cline in performance rather than cursor (Cline being my personal gold standard). $100 is something most people on this forum could make back in 1 day of work.

$100 per month for the value is nothing and for what it’s worth I have tried to hit the usage limit and the only thing that got me close was using their deep research feature. I’ve maxed out Claude code without hitting limits.

Hey, as a maker, I get it. You spent time building something, and you want to understand how it gets used. If you're not collecting personal info, there is nothing wrong with this.

Knee-jerk reactions aren't helpful. Yes, too much tracking is not good, but some tracking is definitely important to improving a product over time and focusing your efforts.

Yes, but increasingly rarely.

I mostly use Perplexity for search, sometimes ChatGPT. Only when I am looking for something _very_ specific do I use a traditional search engine.

Dropping usage of search engines compounded by lack of support led to me cancelling my Kagi subscription and now I just stick with Google in the very rare occasions that I use a search engine at all. For a dozen searches or so a month, it wasn't worth it to keep paying for Kagi.

I’ve built two AI based apps. Earlier this year, I wanted clarity on some issues. I knew several of my books held keys to the answers I was seeking, but finding those felt impossible. I wanted an AI based tool I could use to talk to my books. I built https://www.asklibrary.ai to enable this, implementing tons of RAG tech like query fanout, query understanding and breakdown, multi step retrieval, reranking, etc., and I pull in dozens of pages of text for each answer.

Secondly I love Claude and also use TypingMind but missed the memory feature from ChatGPT. I made MemoryPlugin (https://www.memoryplugin.com) that adds long-term memory to ChatGPT, Claude, Gemini, TypingMind, and LibreChat on desktop and mobile browsers. This got me really interested in AI memory in general, I’ve played around with fine tuning AI models with memories (results = some data learned, way more hallucinations).

Piper seems roughly equivalent to old-school TTS outputs that sound flat, jumpy with the concatenative approach. Listen to this first example I tried:

https://rhasspy.github.io/piper-samples/samples/en/en_GB/ala...

Of all the TTS APIs I have tried, I like OpenAI voices the best. Haven't considered things like elevenlabs because I find them ridiculously expensive.

I love voice to voice interfaces, but only when they sound natural to my ears, and the current pricing for good ones is prohibitive for a huge number of use cases.

Yeah all these seem hyper focused on "voice cloning" so on replicate VoiceCraft doesn't even let you try normal TTS unless you provide a reference voice so I noped out.

From a quick try results aren't good. Sounds bland, and the text I type isn't exactly equal to the text that is spoken. Didn't try with voice cloning though.

Why is good TTS so expensive and why are there no good open source options? Is it just from the need for high quality training data? I don't imagine these models are more expensive to run compared to SOTA LLMs, yet they cost so much more.