Skip to main content
Back

agentmemory hits 11k stars: 92% fewer tokens for Claude/Cursor

Claude Code gets persistent memory, NVIDIA releases Sana diffusion model, Qwen3.6-35B ships free GGUF, HKUDS turns apps into agent CLIs

May 18, 2026 6 min read

Hey,

The real bottleneck in agentic coding isn't model intelligence anymore. It's memory. Every session resets. Every agent forgets.

The open-source community is quietly fixing what the labs haven't prioritized. agentmemory hitting 11k+ stars is a signal, not a coincidence. Developers are tired of being their agent's external hard drive.

We're entering the context engineering era. The winners won't just write better prompts. They'll build better memory.
Together with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 19 sec

Top Repo

▸ Free open-source tool gives Claude Code and Codex persistent memory across sessions

FriendliAI

▸ Optimize inference, grow margins with FriendliAI

Top Repo

▸ Tom Dörr shares 106 Claude Code skills across 15 professions

Kepler

▸ Building deterministic infrastructure for AI

Top Repo

▸ NVIDIA releases Sana, a fast open-source diffusion model for high-resolution images

Signals

1. HKUDS open-sources a tool that turns any software into an AI agent-ready CLI
2. Presented by Buildkite: OpenAI, Anthropic, vLLM, Mistral and Cursor run CI on Buildkite. Orchestration purpose built for reliable scale.
3. Dograh ships open-source, self-hostable voice agent builder with drag-and-drop workflows
4. Qwen3.6-35B mixture-of-experts model gets free quantized GGUF with speculative decoding
Top Repo
Free open-source tool gives Claude Code and Codex persistent memory across sessions
168 Likes
Free open-source tool gives Claude Code and Codex

Your AI coding agent has a memory problem. Every session ends, and it forgets everything. You re-explain your stack, your preferences, your architecture. Every. Single. Time.

agentmemory fixes this. It runs silently in the background, records what your agent does each session, compresses it with AI, and injects the right context when the next session starts. No manual effort. No copy-pasting.

Here is what makes it worth trying:

  • Works with every agent you already use: Claude Code, Codex, Cursor, Gemini CLI, Cline, Windsurf, and more via MCP or REST.
  • 92% fewer tokens per session compared to loading full context, dropping yearly costs to roughly $10.
  • 95.2% retrieval accuracy using a hybrid search that combines keyword matching, vector similarity, and a knowledge graph.
  • Zero external databases needed. SQLite only, fully self-hosted, free.

To get started: run npx @agentmemory/agentmemory, then connect your agent with agentmemory connect claude-code. It has 11.6k GitHub stars and is Apache-2.0 licensed.

Read more
forward →
Presented by FriendliAI
Optimize inference, grow margins with FriendliAI

Latency spikes and GPU bills usually start with the same mistake: benchmarking the easy case.

The Friendli Guide to Inference Performance Optimization shows how to:

  • Mirror real production traffic
  • Measure TTFT, TPOT, E2E latency, and System TPS
  • Tune quantization, speculative decoding, batch size, host KV cache, replicas, and tensor parallelism
  • Choose the cost-efficient point before production finds the breaking point.


1-click deploy 500K+ open models on Friendli Dedicated Endpoints - no infra to stand up.

Get the guide
partner with us →
Top Repo
Tom Dörr shares 106 Claude Code skills across 15 professions
149 Likes
Tom Dörr shares 106 Claude Code skills across 15 p

You know that feeling of typing the same instructions into Claude over and over? That's exactly the problem Claude Skills solve. Think of a skill as a saved playbook: write it once, call it with a slash command, and Claude follows it every time.

Someone just open-sourced 106 of them across 15 professions. Engineering, legal, finance, HR, sales, Figma, marketing, and more. The repo has 324 GitHub stars and is MIT licensed.

The engineering bundle alone covers:

  • Debugging Log Analyser: paste a stack trace, get a root cause diagnosis with a specific fix
  • PR Description Writer: share your diff, get a reviewer-ready description
  • Runbook Writer: deployments, incidents, rollback steps with exact commands
  • Changelog Generator: turns git commits into a polished user-facing changelog

Install the whole library in Claude Code with one command: /plugin marketplace add mohitagw15856/pm-claude-skills. Or grab just the engineering bundle with claude plugin install pm-engineering@pm-claude-skills. The repo claims 8-10 hours saved per week.

Read more
forward →
Presented by Kepler
Building deterministic infrastructure for AI

Past 200k tokens, sourcing discipline erodes. Bigger context windows don’t fix sourced reasoning. They break it sooner. Determinism is the hard part, not selection.

50+ years of Palantir experience, 6 engineers strong, backed by founders of OpenAI, Meta AI, and dbt. Hiring Founding Engineers in NYC.

Read on the blog
partner with us →
Top Repo
NVIDIA releases Sana, a fast open-source diffusion model for high-resolution images
6,222 Stars
NVIDIA releases Sana, a fast open-source diffusion

NVIDIA just dropped SANA, a fully open-source image and video generation framework that runs on your laptop GPU. The big story? It's 20x smaller and 100x faster than Flux-12B, and you can run it with under 8GB of VRAM using 4-bit quantization (shrinking model weights to use less memory).

Here's what the family includes:

  • SANA-Sprint: generates a 1024px image in 0.1 seconds on H100, or 0.3s on an RTX 4090, using one-step generation
  • SANA-Video: beats 14B parameter competitors on video quality benchmarks using only a 2B model, at 36s latency vs 1897s
  • SANA-WM: a 2.6B world model generating 720p, 1-minute videos with full camera movement control
  • Sol-RL: post-training toolkit delivering 4.64x faster convergence for fine-tuning

You can plug it straight into diffusers, ComfyUI, or SGLang today. Clone the repo, run the setup script, and you're generating 4K images in minutes. LoRA fine-tuning, ControlNet, and Cosmos-RL post-training are all supported out of the box.

Read more
forward →
Signals
1
HKUDS open-sources a tool that turns any software into an AI agent-ready CLI
36,016 Stars
2
OpenAI, Anthropic, vLLM, Mistral and Cursor run CI on Buildkite. Orchestration purpose built for reliable scale.
Presented by Buildkite
3
Dograh ships open-source, self-hostable voice agent builder with drag-and-drop workflows
1,882 Stars
4
Qwen3.6-35B mixture-of-experts model gets free quantized GGUF with speculative decoding
5,291 Downloads
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us