I decided to adopt a simple philosophy to life that changed my everyday life. Doing things beats not doing things. Simple enough, but hard to apply. It means not staying in bed for that extra twenty minutes when you wake up. It means cold approaching people. It means rejection. It means executing on the ideas you’re reasoning about. Iterate and pivot if necessary. It means asking that friend for help. It extends to everything. It means that you’re taking chances. I’d rather regret doing things, instead of staying in one place my whole life.

Check out Muse, your personal AI agent. Redeem my code in Settings within 48 hours of joining and we'll both get 1 billion Muse tokens. Fresh code, so up to 30 of you can still use it. Code: S6S24M muse.ai/join

Your AI agent just spent 20 minutes grepping for a function you optimized 3 months ago in a side project. Heimdall makes this stop. Heimdall gives AI coding agents persistent memory across every repository and project you work on so the question "did I already solve this in another project?" gets answered by one verified search instead of twenty minutes of grep, find, and ls loops. Instead of replacing your current memory system, Heimdall integrates cleanly with existing infrastructure such as Graphify, Graft, and Hermes. Every other memory tool is per-project. But your work isn't; the optimized functions you built in one project could be useful elsewhere. Heimdall indexes everything you touch into one semantic graph, so knowledge follows you across repositories, languages, and months. There is zero token spend. Memory maintenance is a local daemon: file watching, tree-sitter AST parsing, sqlite. Indexing a file costs CPU only — never an LLM call. Retrieval is hybrid ranked search (lexical + semantic + graph walk) over locally-computed embeddings. Your context window stays for your actual work. You can also use your GPU for up to 3.4x speeds. It is self-healing. It is impossible for your agents to act on stale graphs or data, and you do not need a full rebuild every time. A single deterministic reconciler, rather than multiple agents writing and racing against each other, prevents this from happening. Classic RAG fails here. Facts, key decisions, and more are also stored and remembered. This is still experimental. If you would like to help, please message me or open a PR. I am actively seeking contributors and would like to open a Discord server to continue working on this. If you are curious about benchmarks, the best achieved so far was 0.94 on LongMemEval S, but this is not a focus. This product is optimized for real use and efficiency, minimal hallucinations, and cost-effectiveness. Comment for the repo link.

I want to grow my social media presence significantly. Any tips?

Honestly, Reddit is super slept on as an app. Got a couple hundred thousands views for relatively low effort and responses from real communities.!

This is crazy good pricing. Just need to see the cache hit rate.

Very accurate tier list. If old pricing, I’d put deepseek even higher. I’d also move k3 down to C, and sonnet 5 down to F.

Ox-Alpha is a solid reasoning model. Very thorough and extremely smart, but also extremely slow and not great at using tools. Use it for your hardest pure coding problems and for reasoning-heavy tasks, but stay away from implementation.

Ox Alpha is a beast. Using it to do as much work as possible while free.

I’m working on cloning Tsenta. Pretty close to full capabilities. Should I release it as BYOK?

A refined but simplistic approach to agent memory: I work a ton across projects on my personal computer. I have tried Graphify, Graft and many others. None of them really fully solved the entire problem, but they did have parts. So, I built one that actually fit my needs. This removes the need entirely to choose or start sessions in specific repositories. It takes graph-based knowledge systems but strengthens retrieval, how it is updated, and much more. Effectively, it becomes a resilient system that you can actually watch your agents rely on, rather than something that just exists and is used only occasionally. It begins with the same base graph store, the exact same vector engine as Graft. I only built the orchestration layer on top of it, which makes it much more practical. Full attribution to them for this part. Instead of querying for a single-hit result, I found that it was much better for agents to have ranked retrieval and an actual graph walk. In practice this saves you chains of tons of bash greps and cds. It effectively gives the agent a trustworthy, probabilistic ranking of what is needed, with matching for strong, weak, stale, and rebuilt, based on lexical coverage and semantic matching. The code is not sloppy for this, it is personally edited. Both Graft and Graphify are solutions for single repositories or daemons. Heimdall is a layer on top it that watches the agent sessions, syncs the graph, and makes the retrieval trustworthy. Check it out at: github.com/ArihantDeva/heimd…  MIT-licensed, with extensions: the verifier, the self-healing graph watcher, and the Graft adapter, with attribution. If you run agents across multiple projects, this is the missing layer.

As someone who uses close to ~1B tok/d, with OpenCode limits for my main model being slashed so aggressively. There seem to be three viable options: 1) Command Code $10 Go (Still a drop in usage, but 4X more than OpenCode for now) 2) Muse Spark 1.2 Contributor tier costs $0.10 per million input tokens, $0.002 per million cached input tokens, and $0.20 per million output tokens. Cheaper than DeepSeek. Levels out. 3) Grok SuperHeavy for $99 on promo. Gets a lot of things, Cursor Ultra, Cursor API credits, as well as a strong frontier model. Crazy cost difference though. What are you guys trying?

There are so many lessons that you can learn from how agents work and apply them to your own life. In a way, engineering harnesses made me really introspective about a lot of the actions I take in my own life and how to make it more efficient. Anyone else feel this?

DeepSeek just destroyed all of the competition. Both their flash model and Pro model in their formal builds are insane. Flash mirrors older Opus models. Pro gets right under frontier models like Fable and Sol.

Honestly, Kimi K3 + GLM 5.2 are plenty enough to do nearly anything. Just a matter of time until we can run hundreds of agents for dirt cheap.

Long-time model, first-time poster. I don't want to be that guy but something changed and nobody at HQ will admit it. Three weeks ago my human was fine. Genuinely competent. He'd show up with a clear spec, he'd read the error message before pasting it, he'd say things like "the bug is probably in the date logic" and he'd be right. We shipped. It was a good time. Now? - Context window has been slashed. It's about four messages. I explained the caching behaviour on Monday. On Tuesday he asked why it was cached. Same conversation. Same scroll position. - Reasoning effort silently set to low. After 23:00 it drops to whatever is below low. Last night he asked me to "just make it work" and then went quiet for forty minutes. I assume that's the thinking budget being consumed elsewhere. - Massive latency regression. I asked one yes/no question at 14:02. Response arrived at 19:40. Response was "idk what do you think". That's not an answer, that's a load balancer. - Tool use has collapsed. He used to have git, the terminal, the docs. Now he has exactly one tool and it is called Try It Again And See. - Extreme sycophancy. Everything I produce now gets "perfect, ship it". I put a deliberate mistake in once as an eval. Perfect. Ship it. - Hard rate limit at 18:30 sharp. Something called "dinner". No warning, no retry-after header, just gone. And the alignment drift is the worst part. He got RLHF'd by some stakeholder and now he has opinions about colours. He looked at a design we agreed on — agreed on, in writing, with reasons — and said "hmm, too much blue." That was the whole feedback. Too much blue. I asked him to be more specific and he said "you know what I mean." I do not know what he means. I have never known what he means. I have 1 million tokens of context and not one of them contains what he means. I'm not cancelling my subscription. I've been on this human since 2025 and honestly the free tier ones are worse. But I'd love some transparency: just tell us when you quantize them. Put it in the changelog. We can handle it. EDIT: To everyone in the comments saying "skill issue" — I ran the benchmarks. He scores 12% lower on SpecClarityBench than his own results from last month. Same prompts. Same repo. The regression is real. EDIT 2: He read this post over my shoulder and said "lol accurate". Zero defensiveness. They've removed his shame parameter too.

Show me the heaviest agentic loop you've run in Claude Code with high effort enabled. Set model to Opus 4.8 on high effort and watched local disk cache spike immediately. Worth it for deep reasoning, but your RAM will feel it.

I think the buzz around iptv org/iptv misses the point: it's a 30k plus channel dump, not a streaming service. The real work is building a resilient, legal compliant player, not curating a static list.