Investor Update – April 2026
Our goal for May is to grow MRR 20% – and we're open minded about if this growth comes from 150 Pro vs 15 Teams subscriptions.
Our goal for May is to grow MRR 20% – and we're open minded about if this growth comes from 150 Pro vs 15 Teams subscriptions.
We were early in making this change, but feel increasing conviction we were correct.
We've added support for package imports and aliases in files.target in shadcn@4.7.0 .
The gap between "works in a demo" and "works at scale" is about 4,000 commits. This is what most of them looked like.
At a million documents, this stops working.
The purpose of this repo (harness and skills) is really to just give the models more direct tools to generate and inspect STEP files.
ERNIE-5.1-Preview builds on the pre-training foundation of ERNIE-5.0 while compressing total parameters to approximately 1/3 and active parameters to approximately 1/2, achieving leading performance at its model scale using only about 6% of the pre-training cost of comparable models.
From the official blog (pageindex.ai), published . View original
It means we've reached a tipping point where most developers can quickly feel at home in Zed.
But the tl;dr is that I actually think we can build a better product, more quickly if we build it with our community + agents.
The VM runs under a locked-down IAM role with zero AWS permissions. It can't touch your infra or ours.
Threads now dock on the left by default, next to the Agent Panel, with the Project Panel and Git Panel on the right.
From the official blog (ernie.baidu.com), published . View original
Browser Use V3 sessions can now use your own LLM API keys (BYOK), with your own rate limits and pricing. A new Code Mode has the agent write a reusable Python script instead of only completing the task once.
Firecrawl v2.9.0 adds the /interact endpoint, so after scraping a page you can click, fill forms or navigate further, either by describing the action in plain language or by writing Playwright code. A new query format on /scrape returns a direct answer to a natural-language question in data.answer. PDF parsing gains fast, auto and ocr modes plus a maxPages limit, and Java and Elixir join the official SDKs.
Drop raw files into a directory; an LLM compiles them into a structured, interlinked wiki of Markdown files.
Session counts and turn counts do not measure outcome quality. A shorter session might produce better code.
Every AI agent needs a browser. We're providing it.
Browser Use Cloud now has a free tier in which every project gets free browser sessions with no credit card required, including scheduled tasks at a minimum one-hour interval. Proxy data now costs $5 per GB, down from $10.
Easydict 2.18.0 adds a MiniMax translation service. Apple TTS now picks premium or enhanced voices over compact ones.
From the official blog (pageindex.ai), published . View original
As more inference providers scale inference using NVFP4 format, this allows Ollama users to share the same results as they would in a production environment.
This is fragile. Every site needs a new script.
Browser Use CLI 2.0 brings Browser Use to the terminal, working with Claude Code, Cursor and other AI coding tools while syncing cookies and sessions from your real browser profile. The first-run flow has been redesigned and includes 5 free tasks with no credit card. On the stealth side, Chromium 146 is deployed, an hCaptcha solver is added and reCAPTCHA accuracy is better, and an update on the page notes that the current CLI is now version 3.0.
The small_model option configures a separate model for lightweight tasks like title generation.
With a codebase approaching a million lines of Rust, mastery isn't expected right away.
From the official blog (lobehub.com), published . View original
From the official blog (lobehub.com), published . View original
From the official blog (lobehub.com), published . View original
AI agents are already browsing the web at scale — searching, comparing products, and making purchase decisions on behalf of users. That shift is only accelerating, and New Generation is building the tools that help brands prepare for it.
The landscape changes, so check what's current when you're getting started.
By default, Zed treats the project search multibuffer like a singleton per pane.
Browser Use has introduced the BU Agent API, an experimental agent built from scratch for web scraping, data extraction, file manipulation and complex multi-step workflows. SDK 3.0 (3.0.x) brings breaking changes and a new client.run() API. Existing features including sessions, profiles, browser control, structured output, streaming, files and skills remain available.
OpenClaw has the ability to read files and execute actions when tools are enabled.
Our split diff view looks like one multibuffer, but under the hood it's actually two.
Ollama now supports subagents and web search in Claude Code. No MCP servers or API keys required.
From the official blog (lobehub.com), published . View original
Change one color, and you're suddenly playing whack-a-mole with readability issues across your entire interface.
It is a Unified Multimodal Model trained from scratch to integrate text, images, video, and audio within a single autoregressive framework.
So we turned to our most valuable resource at Zed: our users.