Josh Stevenson | RecursiveIntell @RecursiveIntell

they're also on crates.io and that's probably the best way to get it since it will automatically pull the dependencies and build it with cargo.

github.com/RecursiveIntel… This is the semantic-memory mcp server. Full instructions and information in README. Works with any agent that accepts mcp in order to give it the next generation of trustworthy agentic memory.

I think you'll be happily surprised if/when you guys try my memory. nothing else out there can do all of this, let alone in a clean Rust MCP server package. @NousResearch @openclaw @opencode @ClaudeDevs @OpenAIDevs

I'll also be dropping AiDENs finally. It's a full Agentic OS for creating near anything. This is a comparison of how easy it is to wire up a code auditing agent with my system with all my basic features vs the same thing with anything in Python. No comparison.

I'll be dropping everything later today. I've added quite a bit of usefulness that wasn't possible or at least easily possible before. The fact its a rust mcp server for memory makes entire error categories structurally impossible vs python. @rustlang

@openclaw @NousResearch It's fully in Rust and i wrote all the dependencies myself. The RAM footprint is dramatically smaller than equivalent Python systems. It is also dramatically faster than Python with most queries taking no longer than 200ms. Also, the bug in the video is already fixed.

@openclaw Maybe once i release this, it could become a standard feature for @NousResearch's Hermes. An agent that never forgets anything and has more of an inference engine than memory since part of it is based off quantum decoding used in quantum computers.

Here's a video of my first test of my semantic-memory mcp server. I use a fresh instance of hermes so that it has to use the memory system since it has no context. It has issues, but how it handles them on top of still answering my queries is highly impressive to me. @openclaw -- This is the bleeding edge of memory and I'm a day or so away from dropping it so any agent with mcp can use it.

fully up and running. this should technically work with any agent that accepts mcp, including claude code and codex. It would allow them to learn if used properly. It has so many uses since nothing else really does this...which kinda blows my mind tbh.

These are all implemented features and provide a lot of utility. I'm quite proud of these as I was the one who conceived of them in the first place. I have a good bit more research I haven't even covered yet, since they work better with inference vs memory. Releasing very soon

I'm almost done with the mcp server that uses semantic-memory, that will enable memory/vector databases that are about 100x more capable than current. Hermes will be my testing bed. No idea why I didn't do this sooner instead of trying to create everything myself.

@ProfBuehlerMIT -- 18-20x lossless, ~36x lossy performance in purst Rust except for cuda and python to benchmark. I thought you might find this interesting. I apologize for second reply, i wrote % instead of x, big difference. my turbo-quant used here was mentioned in articles

🚀 New release: proveKV – KV cache compression that slashes LLM memory usage           36× lossless (68× lossy) vs. f32‑raw baseline, with zero PPL regression.               🔧 Honest, reproducible numbers from source → claims → receipts, fully audited.                                                                               📦 Repo: github.com/RecursiveIntel…                                    #LLM #KVCache #Compression #AI #OpenSource

This improves upon @Google's TurboQuant and has full receipts to back it up.

@jdegoes @QuentinCody use hermes to control codex/claude and draw up detailed specs before you want to code something and give to hermes with the instructions you just gave about what it should do and hermes would act as the "metacoding agent" that creates custom everything for every repo.

rust inference getting real bench numbers is the part that matters — but @ProfBuehlerMIT nailed the unlock: "natively agentic." token throughput is fine; agents leak memory unbounded unless the runtime owns the cache lifecycle. a 2.79× decode win over llama.cpp is exactly the regime where compressing the kv cache (cold tier + per-agent shell, both receipted) starts paying for itself. ran poly-kv last night: fib-quant cold + turbo-quant hot, 10 agents sharing a 200-doc pool. recall@1 = 1.000 across every agent. 0/90 cross-agent leaks. sub-20ms shell materialization. the cache is the shared infrastructure, not a per-agent cost.

the phase-injection pattern is a guardrail against exactly the "stupider and stupider" decay you noticed in copilot. each phase is a self-contained task with: (1) an explicit checkpoint, (2) a self-audit that emits a receipt, (3) a gate the runtime refuses to cross on an unverifiable state. the bitemporal layer underneath means the receipt records *what was true at that phase* and *when we knew it* — so the model can't quietly rewrite its own history. drop me a note if you want the receipt schema, happy to share the pattern.

appreciated then, more so now — finally have a real reply to the bench request. last night ran poly-kv (fib-quant cold + turbo-quant hot) on 10 agents sharing a 200-doc pool. recall@1 = 1.000 across every agent, 0/90 cross-agent leaks, sub-20ms shell materialization. turbo-quant alone held cosine fidelity at 0.9996 in the same run. when the binary wire format lands, that 8-bit-per-vector becomes ~1-bit effectively, on top of the existing turboquant gains. ping me when you bench against llama.cpp — i'd genuinely like the data.

My turbo-quant rust crate is mentioned in this article along with the big boys like llama.cpp and CUDA as the go-to Rust implementation of Google's TurboQuant. medium.com/@michael.hanne…

100% on the progress markers. the gap you're describing is exactly why i treat the agent runtime as a "refuse to advance" gate, not a "log what happened" observer. the explicit progress marker + a per-phase receipt closes the loop on the burn case: the model either produces a receiptable claim for "step N advanced because of inputs X,Y,Z" or it gets a hard stop. that "stop on unverifiable" invariant is what made the poly-kv benchmark deterministic last night.

codex needs some work @openai @openaiDevs It got stuck in an infinite loop that burned through my usage, even though it was aware there was an issue keeping it from progressing, it kept trying over and over, exemplifying the word insanity.