8. mem-agent mem-agent is a 4B-parameter LLM trained with GSPO reinforcement learning to develop persistent memory using a scaffold of Python tools and markdown files.
CODE
-
Markovian Thinker: RL Environment for LLM State Management
By
–
6. The Markovian Thinker
— DAIR.AI (@dair_ai) 12 octobre 2025
A new RL thinking environment that keeps an LLM’s effective state constant by chunking long chains of thought and carrying over only a short textual state between chunks.https://t.co/mqXfF1XG6h6. The Markovian Thinker A new RL thinking environment that keeps an LLM’s effective state constant by chunking long chains of thought and carrying over only a short textual state between chunks.
-
Building More Efficient AI Agents for Specialized Tasks
By
–
i made a chatgpt assistant "Fix Anything". It can fix anything, but can't fix the cancer for my 11 year old son. If you can help make a more efficient assistant using Agent AI. Thanks having credits api.
-
IDE-based AI chat tool for improved diff visualization workflow
By
–
It’s been a big part of my workflow in the past few weeks – Looking forward to where it’s headed I only wish I could have better IDE based chat w/ it too primarily to visualise the diff better!
-
AI Programming: Balancing Automation with Manual Work
By
–
while programming with AI, hard to decide whether to be inspired by how much it does, or annoyed by any manual work left for you to do
-
DeepAgent: Powerful Coding Agent Now Available for Browser Use
By
–
DeepAgent Coding Agent Using The Browser
— Abacus.AI (@abacusai) 12 octobre 2025
We now have a powerful coding agent that can be directly used on the browse and it’s exceptionally powerful pic.twitter.com/GKNi5zqN8VDeepAgent Coding Agent Using The Browser We now have a powerful coding agent that can be directly used on the browse and it’s exceptionally powerful
-
SDK Auto-Compaction and Token Usage Statistics
By
–
We auto-compact for you when using the SDK, so you don't need to. To grab stats yourself, read assistant_message.usage.
-
Rate Limits and Token Compaction Bugs in API Systems
By
–
If you're seeing lower rate limits than what we publish, or if you're seeing auto-compact earlier than ~155k tokens, that's a bug. /bug to report.
-
Claude Code Team Updates Context Token Auto-Compact Behavior
By
–
Boris from the Claude Code team here. Compact behavior is the same as before — the new ⛝ boxes in /context are just a cosmetic UI change that gives people more transparency into auto-compact. We always auto-compacted near 155k tokens so there's enough buffer. We do that for
-
PyTorch Multiprocessing Tensor Sharing Across Processes
By
–
pytorch uses this for sharing tensors across processes in torch.multiprocess