[Dev Catch Up # 106] - Anthropic's Mythos, Meta's Muse Spark, Amazon's S3 Files, llmfit, OpenClaw's Dreaming feature,Claude Managed Agents,Graphify, nono-sandbox,Perplexity's plaid integration & more!
Bringing devs up to speed on the latest dev news from the trends including, a bunch of exciting developments and articles
Welcome to the 106th edition of DevShorts, Dev Catch Up.
For those who joined recently or are reading Dev Catch Up for the first time, I write about developer stories and open source, partly based on my work and experience interacting with people all over the globe.
Thanks for reading Dev Shorts! Subscribe for free to receive new posts and support my work.
Some recent issues from Dev Catch up:
Join 8800+ developers to hear stories from Open source and technology.
Must Read
Anthropic’s new Project Glasswing is getting a lot of attention. It is a new effort to secure critical software. It is built around Claude Mythos Preview, Anthropic’s latest model. Anthropic says Mythos can find serious software flaws. It has already identified flaws in the Linux kernel and major browsers. Because of its capabilities, Anthropic is not making it publicly available. Check Anthropic’s announcement for more details.
Meta is back in the model race with Muse Spark. It is the first model from Meta Superintelligence Labs. It supports multimodal reasoning and visual chain of thought, and multi agent tasks. Check Meta’s announcement for more details.
Amazon has announced S3 Files. It lets you access S3 buckets like a file system, without moving data out of S3. It gives file system access to S3 data with low latency. Check AWS’s announcement for more details.
OpenClaw has released a new memory feature called Dreaming. It is a background memory consolidation system for long term recall. It takes useful short term context and stores it in durable memory. It is off by default. Check OpenClaw’s docs for more details.
OSS Highlight of the Week
This week we are featuring nono. It is an open source sandbox for AI agents. It lets you run agents in an isolated environment without extra infrastructure. It also supports tools like Claude Code, Codex, OpenCode, and OpenClaw. Check the nono GitHub repo for more details.
Good to know
Anthropic has launched Claude Managed Agents. It helps teams build and run cloud hosted agents on Anthropic’s infrastructure. Anthropic handles things like sandboxing, permissions, state, and tracing. So teams can focus more on the agent itself. Check Anthropic’s announcement for more details.
LangChain has introduced Deep Agents Deploy. It is an open alternative to Claude Managed Agents. It helps teams deploy production ready agents with one command. It is model agnostic and open source, so teams are not locked into one provider. Check LangChain’s announcement for more details.
If you keep a lot of notes in markdown, Atomic is worth a look. It is an open source personal knowledge base. It turns markdown notes into a connected knowledge graph. You can explore it through wiki style pages, a canvas, and chat. Check the Atomic GitHub repo for more details.
xAI looks to be scaling up fast. Elon Musk says they have 7 models in training. They include Imagine V2, two 1T variants, two 1.5T variants, a 6T model, and a 10T model. Check Elon Musk’s post for more details.
IBM Research has released ALTK Evolve for AI agents. It helps agents learn from past runs instead of just re reading old transcripts. IBM says it turns agent traces into reusable guidelines, and improves reliability on harder tasks. Check the Hugging Face post for more details.
Microsoft has released Memento. It is a new way to extend LLM output length without needing a bigger context window. It breaks reasoning into blocks, writes short summaries, and clears old KV cache. So the model can keep going longer. Check the Memento GitHub repo for more details.
Notable FYIs
llmfit looks useful if you run models locally. It checks your RAM, CPU, and GPU, then tells you which models will run well on your machine. It also scores them across fit, speed, quality, and context. It comes with both a terminal UI and a CLI mode. Check the llmfit GitHub repo for more details.
The skill called Graphify caught my attention. It is an open source skill for Claude Code and other coding assistants. It turns a folder of code, PDFs, markdown, and images into a queryable knowledge graph with one command. It can help you understand a codebase faster and trace architecture decisions. Check the Graphify GitHub repo for more details.
Claude Code has added /autofix-pr in the CLI. You can now trigger PR autofix directly after finishing your PR. It sends your session to the cloud, so the autofixer gets full context to handle CI failures and review comments. Check the X post for more details.
Perplexity has added Plaid integration to its Computer agent. Users can now link bank accounts, credit cards, and loans for a fuller view of their finances. Perplexity says Computer can analyze spending, track net worth, and build custom finance tools from simple prompts. Check Perplexity’s announcement for more details.
That’s it from us with this edition. We hope you are going away with a ton of new information. Lastly, share this newsletter with your colleagues and pals if you find it valuable. A subscription to the newsletter will be awesome if you are reading it for the first time.