Dwarves Foundation
@dwarvesf
Followers
2K
Following
83
Media
485
Statuses
1K
Dwarves build and ship top-notch software for tech-focused companies across the globe since 2015. Discord: https://t.co/fzlq3mTsRX 💻🎧☕
Internet
Joined January 2015
📚Built on @0xPolygon, by our own engineers, $ICY takes co-creation within @dwarvesf Network to the next level. Join us, engage with us, learn with us, build with us, nurture the community with us, and get $ICY in return. 📌About ICY: https://t.co/tmO9Vq0iCr
2
4
8
People run into this curve all the time. At some point you add scripts, helpers and small agents around it because it feels like progress. Then you realise you spend more time keeping that stack alive than shipping. Image credit goes to @thorstenball
1
1
2
Stagehand nails the middle ground: Playwright for deterministic steps + intent-based actions when UI shifts. A11y-tree targets > CSS. observe→preview, then act; extract(schema) for typed data; cache & replay. Full breakdown by Chinh Le: https://t.co/erKCHBuLPz
We've built the best in house observability for Stagehand 🔭 Get inference time, token usage, LLM output, and more on the new dashboard. No more black box, see everything behind the scenes. See for yourself, get started with npx create-browser-app 🤘
0
1
1
By Lap Nguyen (Frontend @ Dwarves): E2B explained. Firecracker microVMs run untrusted agent code with VM-level isolation and fast cold starts. Covers architecture, lifecycle, snapshots, quotas, scaling. Check out: https://t.co/4Nfc0oXyTI
0
1
4
Breakdown by Quang Huy (Backend @ Dwarves): Maybe Finance for self-hosted personal finance. Keys: Family as tenant root, delegated Account types, append-only Entry + anchor balances for correctness; 3-tier cache; SQL-level LOCF. Details:
memo.d.foundation
An in-depth analysis of a $1M open-source personal finance application built with Ruby on Rails
I have a breakdown for Maybe https://t.co/l2GgsiCmUp
0
3
4
Agentic build on @dify_ai: tools with schemas (HTTP/DB). Loop: React or function calling + HITL checkpoints. RAG: hybrid + parent-child; chunk600, overlap100, top-k~8. Streaming on; retry with backoff. SSRF-safe proxy. DifySandbox. https://t.co/s5RwRppGuP by @_zlatanpham
📚 Build Deep Research Agent: From Question to Insight Complex research rarely fits in one search box. The Deep Research workflow built with Dify Iterates, reasons, and sources until the answer is complete. 💡 Key workflow nodes include: 1) Start: capture the research topic
0
0
2
Track traffic without GA4 or cookie banners and keep data ownership. Umami is a self-hosted, privacy-first alternative focused on minimal setup and end-to-end control. Breakdown by Chinh Le (our frontend engineer): https://t.co/Toyk1E6ETP
Umami is now the most starred open source analytics project! And we launched later than everyone else 😎 A huge thanks to our community for making it happen. More great things to come! 🥳 #opensource #Analytics #GitHub
0
0
4
MEM0: Modular memory for agentic AI Decouples memory from inference. Graph-based reasoning, implicit forgetting, multi-backend storage. +26% acc, -91% latency, -90% tokens vs OpenAI Memory. Mem0 breakdown was composed by @minhlq96. Check out the link below.
Just making a breakdown about this great memory layer here
0
2
8
Crawl4AI: clean Markdown (variants + citations) + typed JSON for RAG/agents. Pick speed (CSS/XPath/Regex) or accuracy (LLM schema). Anti-bot ready with Playwright + proxies; scales with browser pools + cache. Our takes in the breakdown series: https://t.co/BUQUXl8n41
Crawl4ai crossed 50K start 🌟 (~5k forks). Thanks to all who shared their love and support. A project I started to build a tool for my company turned into a lovable open-source project at this scale. It's so rewarding to be part of an authentic change. I like to believe Crawl4ai
1
2
11
Cline brings structured tool calls, streaming diffs, and shadow commits to VS Code. We put together a developer-first breakdown, how it works, where it shines. Written by Chinh Le (frontend @ Dwarves): https://t.co/hBWEoKLLmb
0
0
1
ax framework is a TypeScript port of @DSPyOSS. Instead of writing prompts directly, you can pass arguments to the framework and it will generate the prompts, which makes managing the semantics of the codebase a bit easier. This piece is a part of the team’s breakdown series.
https://t.co/kORIQedGAq one of my favorite tool when building agentic apps. Read this if you're still on the fence about adopting DSPy.
0
0
3
Why care? - No more manual state management. - Context sticks around, no matter how many tools you use. - Workflows stay sharp and efficient.
0
0
0
Use Cases - Seamless cross-tool investigations. - Reviving context after a reset. - Getting consensus from different models.
1
0
0
Tool Categories - Simple tools: chat, challenge, listmodels, version. - Workflow tools: analyze, debug, code review, consensus, planner, secaudit, and more. - Special tools: tracer for code flow, hybrid challenge for critical thinking.
1
0
0
Core Capabilities - Persistent conversation memory, even after resets. - Cross-tool continuation to keep context flowing. - Dual prioritization keeps the right files and history in play. - Multi-provider routing always picks the best model for the job.
1
0
0
What’s Zen MCP? - Converts stateless AI requests into threads with memory. - Bridges context between tools and providers. - Orchestrates workflows across Gemini, OpenAI, local models, and more.
1
0
1
Most AI tools just stack models, but context and collaboration still get lost. zen-mcp-server by @BusyMac gives Claude real multi-model memory, so your AI team actually works together, not in silos. Here’s how we break it down at Dwarves:
memo.d.foundation
Technical analysis of the Zen MCP (Model Context Protocol) Server architecture, implementation, and design patterns.
Somebody hooked up Claude Code to pair program with o3 and Gemini 2.5 via MCP for an all-star coding lineup 🤯 - Claude controls the work - Can call out to Gemini and o3 for tasks and input - Works around MCP limit by using prompt files Check out Zen MCP (link below)
1
2
9
Meanwhile, I’m out here hitting Claude’s usage cap 3 times a day… and still not on the leaderboard. We are, indeed, absolutely r̵i̵g̵h̵t̵ cooked. 👍
Some of the biggest Claude Code fans are running it continuously in the background, 24/7. These uses are remarkable and we want to enable them. But a few outlying cases are very costly to support. For example, one user consumed tens of thousands in model usage on a $200 plan.
0
0
0
Claude: I converted all 106 models flawlessly. Reality: Literally none of them work. AI delusion speedrun (any%) complete. #devlife #techmemes #AIart
0
0
1
Bro really took the MBTI test and said ‘I’m not like other browsers’. #comet #Perplexity
0
0
0