Tools Bench.

Product launches and open-source repos with enough signal to earn a second look.

Last Brew Time: Jul 18, 2026, 10:41 AM PT

Insight

This run's sharpest builders are patching the seams where agents actually break

01.FEEDING CODING AGENTS REAL CODEBASE CONTEXT

02.CATCHING THEIR BUGS BEFORE A HUMAN HAS TO

03.PUTTING A LEASH ON WHAT THEY'RE ALLOWED TO TOUCH

Featured

GitHub36.5K

PostHog/posthog

🦔 PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

youtube
What is PostHog? (Official Demo & Tutorial)

Market Signal

Why It Has Market Pull

PostHog is a mature, thriving open-source product analytics and AI observability platform with a large paying customer base and continued rapid revenue growth. Its scale and current AI-focused feature velocity make it a legitimately strong signal even though it is not a new project.

  • 36,509 verified GitHub stars and 3,023 forks as of July 2026; pushed the same day this data was pulled, reflecting an extremely active engineering org
  • Raised a $75M Series E led by Peak XV at roughly a $1.4B valuation (late 2025), on top of $194M total raised across 7 rounds
  • Reported ~$57.5M ARR in February 2026, up ~99% year-over-year
  • 4,897 open GitHub issues — evidence of a very large, active user and contributor base rather than an abandoned tracker
  • Active, ongoing build-out of AI-specific product surfaces (signals, autoresearch, pulse) visible directly in recent commit/issue history

feedbacks

What People Are Saying

  • "PostHog built a platform that bundles product analytics with session replay, feature flags, A/B testing, surveys, error tracking, and data warehousing"Vibe Coder Blog comparison

  • "best for engineering-led teams wanting analytics, session replay, feature flags, experiments, error tracking, LLM observability, and surveys in one place"userpilot.com comparison

  • "$57.5M in annual recurring revenue... up approximately 99% year-over-year"Sacra research

  • "raised a $75M Series E led by Peak XV"funding press coverage

  • "feat(autoresearch): add product skeleton, training loop, MCP tools, and admin"GitHub issue

  • "feat(pulse): add product skeleton and on-demand brief pipeline"GitHub issue

  • "feat(signals): scout fleet + per-scout scheduling"GitHub issue

GitHub20.1K

tirth8205/code-review-graph

Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.

Market Signal

Why It Has Market Pull

code-review-graph is a local-first, Tree-sitter-based code intelligence graph exposed over MCP/CLI so AI coding assistants read only the relevant slice of a codebase. It has scaled to roughly 20k stars in under five months with a near-daily release cadence and a genuinely broad, active contributor base rather than a single maintainer.

  • 20,093 verified GitHub stars and 2,123 forks as of July 2026, up from a February 2026 launch — five months old
  • Near-daily shipping cadence: v2.3.7 released July 18, 2026, with prior releases in June and May 2026
  • Roughly 98 distinct contributors beyond the primary maintainer, plus 665 issues opened to date showing sustained engineering activity
  • Listed across multiple MCP directories (mcp.so, Glama, mcpmarket, cursor.directory) and passed an independent SkillsLLM security scan with no high-severity findings
  • Featured in a Hacker News submission on cutting AI coding-assistant token usage via a persistent code graph

feedbacks

What People Are Saying

  • "27,700+ files excluded from review context, only ~15 files actually read"SkillsLLM coverage

  • "How I Set Up code-review-graph on My Spring Boot Project with Cursor — And Why It Changed How I Review Code"Medium article

  • "passed SkillsLLM's automated security scan... with no high-severity issues"SkillsLLM coverage

  • "fix(platform): validate Copilot and Antigravity ports"GitHub issue

  • "fix(parser): distinguish C++ overload identities"GitHub issue

  • "fix(search): preserve cross-repository result visibility"GitHub issue

GitHub12.7K

Robbyant/lingbot-map

A feed-forward 3D foundation model for reconstructing scenes from streaming data

youtube
LingBot-Map: New 3D Model for Fast Reconstruction

Market Signal

Why It Has Market Pull

LingBot-Map is a streaming 3D scene-reconstruction foundation model open-sourced by Robbyant, the embodied-AI unit of Ant Group, in April 2026. It has drawn real engineering engagement rather than pure hype, and ships with an arXiv technical report plus Hugging Face/ModelScope weights.

  • 12,747 verified GitHub stars and 1,332 forks as of July 2026, roughly three months after its April 2026 launch
  • Backed by Robbyant / Ant Group, with independent coverage from Business Wire, Yahoo Finance, and Morningstar
  • 88 GitHub issues opened to date covering real technical topics (Apple Silicon support, GPU memory tuning, COLMAP/NeRFStudio integration), most recent July 17, 2026
  • Ships a full technical report on arXiv plus released model weights on Hugging Face and ModelScope, with published benchmark results (e.g. 6.42m ATE on Oxford Spires)
  • Actively pushed as recently as July 12, 2026

feedbacks

What People Are Saying

  • "Add Apple Silicon (MPS) support with bf16 autocast"GitHub issue

  • "Guide to render new camera angle"GitHub issue

  • "Any plans for loop closure and global pose graph optimization?"GitHub issue

  • "Min memory (2.05GB), higher FPS (1.6x), balanced mode — pick whichever you like."GitHub issue

  • "3D Reconstruction - COLMAP & NerfStudio integration"GitHub issue

  • "operates on a 'see-as-you-go' principle, continuously estimating camera position frame-by-frame"Business Wire coverage

Product Hunt584

Unabyss for Claude

Claude doesn't know what happens in GPT. Neither one really knows who you are or what your company does. Now they can. Unabyss gives Claude memories from your other AI agents and everyday apps: email, Drive, GitHub, Notion, meeting recorders, and 20+ more. It saves new memories too, so GPT and Cursor stay in sync with the exact same context - sharper than wiring each tool into Claude one by one. Finally, a real memory that follows you. Private. Portable.

youtube
Unabyss recensione severa: il context layer MCP che si auto-aggiorna vale davvero?

Market Signal

Why It Has Market Pull

Unabyss is a personal AI context layer — a structured vault of a user's identity, preferences, and knowledge, pulled once from sources like email, Drive, GitHub, Notion, and LinkedIn, that any connected AI tool (Claude, ChatGPT, Cursor, and others) can read and write to via MCP, so context follows the person instead of staying trapped in one app. It has now hit #1 Product of the Day on Product Hunt twice.

  • Hit #1 Product of the Day on Product Hunt on two separate launches (most recently with 584 upvotes and over 130 comments)
  • Backed by institutional seed investors including Kogito Ventures, a Warsaw-based early-stage VC fund
  • Grew to 2,400+ Product Hunt followers between its two launches
  • Launch-day discussion from technically sophisticated users focused on a real risk of shared memory — stale facts propagating across agents — and the team responded with source/repetition/authorship-weighted conflict resolution plus a roadmap for isolated memory silos
  • New users get $20 in free usage credits, then pay-as-you-go pricing

feedbacks

What People Are Saying

  • "the file being a byproduct instead of a chore"Product Hunt comment

  • "the moment context is a file you maintain, it drifts, because keeping four copies honest is a chore"Product Hunt maker reply

  • "how do you handle memory conflicts when different tools have slightly different or outdated context?"Product Hunt comment

  • "Hit this exact thing jumping between Claude and GPT on the same project. Half the context just vanishes"Product Hunt comment

  • "maintained CLAUDE.md files across client projects...Mine rot quietly until something breaks"Product Hunt comment

  • "recency as the tiebreaker assumes newer means truer, but a stable preference from 3 months back usually beats something I typed once yesterday"Product Hunt comment

  • "one agent writes a fact that's slightly wrong or goes stale, and every other agent confidently inherits it"Product Hunt comment

Product Hunt427

Kimi K3

Kimi K3 is the world's first open 3T-class model — frontier performance across coding, knowledge work, and reasoning, with native multimodality and 1M context.

youtube
Kimi K3 Is Fable Level... (they should be worried)

Market Signal

Why It Has Market Pull

Kimi K3 is Moonshot AI's newest flagship model — a 2.8-trillion-parameter open-weight mixture-of-experts model with a 1-million-token context window, described as the largest open-weight model released to date. Early independent benchmarks and hands-on testing place it near the top tier of frontier models on coding and long-context tasks, trailing only the very best closed models, and full open weights are scheduled for release under a permissive license. This listing is a re-launch of the same release and drew even stronger community response than the first.

  • This re-launch drew 427 upvotes on Product Hunt, far above the 55 upvotes on the original listing, reflecting sustained interest
  • Backed by roughly $3.9B in funding to Moonshot AI over the prior six months, at a reported $20B valuation, with Alibaba and Tencent among the investors
  • Covered by major press (Bloomberg, Axios, Fortune, Tom's Hardware) as a frontier-level open-weight release that narrows the gap with leading US labs
  • Independent testing (Simon Willison) rated its vision/image-understanding output as strong, while noting reasoning-token costs run high for simple tasks
  • Full open weights (Modified-MIT license) scheduled for release days after the initial launch, enabling self-hosting

feedbacks

What People Are Saying

  • "Vision works well: the alt text it generated is very good."Simon Willison blog

  • "a simple task cost 25 cents due to mandatory maximum reasoning effort"Simon Willison blog

  • "couldn't tell it apart from Fable on real coding work"Hacker News comment

  • "Moonshot Unveils Kimi K3 AI Model, Narrowing Gap With US Rivals"Bloomberg

  • "beats Claude Fable 5 in Frontend Code Arena benchmark"Tom's Hardware

  • "China's open-weight Kimi model stuns AI world with frontier-level results"Axios

Sources

GitHub

apache/ossie

1.2K

Apache Ossie, industry wide specification effort to standardize how we exchange semantic metadata across analytics, AI and BI platforms, providing a vendor neutral, single source of truth for semantic data

Product Hunt

Pebbles Ai

426

The only GTM orchestration platform you will need to successfully take your products & services to market. Pebbles AI is a Go-To-Market Operating System built for B2B revenue teams. It brings strategy, lead generation, outreach, sales, & shared company knowledge into one AI-powered workspace. Using neurosymbolic AI trained on your business, it helps teams plan campaigns, personalize outreach, generate qualified leads, & execute without switching between disconnected GTM tools.

Basedash Suggestions

155

Basedash now suggests the analysis before you ask. It studies your connected data, your past chats, and the dashboards you've built, then generates personalized suggestions — questions worth asking, dashboards worth building, automations worth scheduling. Click one and the work starts. Used ideas are replaced with fresh ones, so the well never runs dry. Every suggestion is generated per person, for growth, finance, and ops alike. No more blank page. Your analyst makes the first move.

Aye

153

Aye is a Chromium-based AI browser for macOS and Windows that gives web work a teachable AI intern. It reads visible pages, plans steps, and works through normal browser actions: clicking, typing, scrolling, switching tabs, and checking results. Summarize pages, research across tabs, draft replies, and automate repeatable workflows. Turn recurring tasks into reusable skills, separate accounts with profiles, and stay in control with reviewable progress and approval for sensitive steps.

Kimi K3

55

Kimi K3 is a 2.8T-parameter open model featuring native vision capabilities, a 1-million-token context window, and Moonshot AI's Kimi Delta Attention and Attention Residuals architectures. Built as the world's first open 3T-class model, it delivers frontier-level performance in long-horizon coding, compiler development, digital creation, and scientific reasoning, outperforming previous open models in scaling efficiency and agentic capabilities.

Pocket Screen

241

Pocket Screen turns the frontmost window on your Mac into a compact, always-on-top PiP-style view. Keep documents, chats, videos, or reference material visible while you work in another app—without constantly switching windows. Processing stays on your Mac.

Timely

191

Pull your availability from connected calendars, with a single keystroke, within whichever timezone suits your recipient best. No opening the calendar, no timezone maths, and it handles multiple execs too. It's completely private and sits locally on your machine, connecting only to Outlook and Google.

YC Launch

Rational: AI Employees for Accounting

12

24/7 hires to solve the talent shortage within your firm's existing stack Rational · Summer 2026 · B2B Tags: Enterprise, AI. Website: https://rational.to

Alkera - The data agent that you can trust

8

Alkera is an agent that replaces your coding agent to reliably and safely perform real data engineering/analysis/science work. Alkera AI · Summer 2026 · B2B Tags: Artificial Intelligence, Developer Tools, Data Science, Data Engineering, Enterprise Software. Website: https://alkera.ai

Hacker News

FixBugs – Reproduce production bugs and verify fixes

42 pts

I built FixBugs, an agent that ingests the rich context surrounding production bugs to reproduce them in a sandbox and generate verified fixes. It's available in the form of a self-hosted VSCode extension and as a Github app: VSCode Extension: https://fixbugs.ai/go/vscode-extension - full code and data privacy. - zero data retention models opted out of training. GitHub App: https://fixbugs.ai/go/github-app - we do access your code temporarily. - pick a repo to install FixBugs on. What motivated... (42 points, 39 comments).

Oodle.ai – $10 per million agent traces

31 pts

Hi HN, we're Kiran and Vijay! Over the past two years, we have built a columnar storage engine for observability: logs, metrics, and traces. Today, it's exciting for us to show what we've built on top of that foundation: LLM Agent Observability. Given how non-deterministic agents are, storing all traces without sampling was critical for us. But these traces tend to be in the MBs, sometimes GBs - we needed to store them inexpensively. We also needed the queries and analyses to be fast. To meet bo... (31 points, 12 comments).

Ratel, give agents unlimited tools and skills without context bloat

23 pts

Hi HN! We're Giacomo and Roberto, authors of Ratel ( https://github.com/ratel-ai/ratel ) We used to help SaaS companies build agents on top of their products. Whenever we wanted to expand the agents’ complexity/scope, by adding more and more tools and instructions, we always run in the same issue: context bloat, with frequent hallucinations and sky high token bills. So we started constantly engineering the agents, dynamically loading tools, splitting them into subagents, inventing our own way to... (23 points, 18 comments).

Policy enforcement for Claude Code, Cursor, and Codex

13 pts

Show HN: Runtime authorization for Claude Code, Cursor, and Codex Hi HN, Fernando and I built Kastra. Kastra intercepts AI agent tool calls and evaluates them against deterministic policies before they execute. This is aimed at developers using coding agents like Claude Code, Codex, Cursor, and OpenClaw. We built Kastra after one of our Cursor agents almost executed DELETE FROM customers WHERE status='test' against a production database. We caught it before it ran, but it made us realize that no... (13 points, 5 comments).

I hated how much my 12-year-old played Roblox, so we built our own FPS

23 pts

I'm a father of two, 7 and 12. They are obsessed with Roblox, especially Rivals. Like a lot of parents, we did not love it. We tried the usual things: block it, limit it, set timers." It became a daily battle, a lose-lose situation. So I flipped the problem. Instead of fighting what they loved, I decided to lean into it, but with a twist. Why just play an FPS when you could build one together? My kids became the PMs. Claude and I became their engineer. I was shocked by how fast we moved. We pick... (23 points, 7 comments).

Libretto PR agents – Automatically fix failing playwright scripts

21 pts

Libretto PR agents is a free TypeScript library for maintaining Playwright browser automations. Add one line of code to your existing Playwright scripts and it lets an agent automatically open GitHub PRs fixing the script when it fails. A few months ago we released Libretto, a CLI + coding-agent skill for building deterministic browser automations. The idea was that for many browser workflows, especially repetitive business workflows, you don’t need an AI agent making decisions at runtime. You w... (21 points, 4 comments).

HF Spaces

DeepSite v4

16.6K likes

Generate any application by Vibe Coding it DeepSite is a Vibe Coding Platform designed to make coding smarter and more efficient. Tailored for developers, data scientists, and AI engineers, it integrates generative AI into your coding projects to enhance creativity and productivity. DeepSite v4 is a Hugging Face Space tagged with docker, region:us. It has 16617 likes on Hugging Face.

Kolors Virtual Try-On

10.1K likes

Kolors Virtual Try-On is a Hugging Face Space tagged with gradio, region:us. It has 10129 likes on Hugging Face.

Wan2.2 Animate

5.1K likes

Wan2.2 Animate is a Hugging Face Space tagged with gradio, region:us. It has 5116 likes on Hugging Face.

Arena Leaderboard

4.9K likes

Arena Leaderboard is a Hugging Face Space tagged with static, leaderboard, region:us. It has 4946 likes on Hugging Face.

The Ultra-Scale Playbook

3.9K likes

The ultimate guide to training LLM on large GPU Clusters Instruction to install and run locally Loading HTML fragments: There are two way to load HTML fragments: Compile them into html during build time Fetch them and insert them during run-time When to use what Use compile time fragments only on parts which you want to ensure are seen by every user right after page load (e.g logo) Use run-time fragments for everything else so that the final HTML is of reasonable size (<1MB idealy) How to add a new fragment Add it to the src/fragments folder (e.g. src/fragments/banner.html) For run-time fragments, add {{{fragment-name}}} to appropriate place in src/index.html (e.g. {{{fragment-banner}}}) For...

Live Portrait

3.8K likes

Apply the motion of a video on a portrait Live Portrait is a Hugging Face Space tagged with gradio, Multimodal, Motion control, Image-to-Video, Video-to-Video. It has 3763 likes on Hugging Face.

AI Tools — July 18, 2026 Edition | Agentic Brew