Insights

Data-driven landscape analyses of the AI open-source ecosystem. Every metric updated daily from 247,805 tracked repositories.

Your Agent Framework Is a Ball of Mud — Who's Actually Fixing It

You've lost sessions to compaction, watched agents redo work, and restarted after crashes with nothing to resume from. Claude Code's leaked source reveals why: brilliant simplicity designed for interactive work, hitting an architectural ceiling as agents go autonomous. Here's who's building what comes next.

Graham Rowe · April 05, 2026 · agents rag data-engineering mlops

Your Docs Are Written for Humans. Your Users Are Agents.

Agents choose developer tools based on documentation quality. llms.txt is the new robots.txt. The MCP data proves it: repos with detailed docs get 18x the stars.

Graham Rowe · April 03, 2026 · mcp agents llm-tools

Your Agent Doesn't Have an Email Address (Yet)

30+ repos are building identity, credentials, email, and payment infrastructure for agents as first-class entities. Scattered across 25 subcategories with no name. Here's the map.

Graham Rowe · April 03, 2026 · agents mcp

Your Newest Competitor Isn't a Developer

27% of AI repos created in 2026 sustain 300+ commits per month. New GitHub accounts, first repos ever, absurd velocity. Here's what's happening — and what it means for the developer tools market.

Graham Rowe · April 03, 2026 · ai-coding agents ml-frameworks

You're Shipping AI You Can't Measure

1,159 repos are building LLM evaluation infrastructure. Most teams are still eyeballing outputs. Here's the decision guide to what actually works.

Graham Rowe · April 03, 2026 · llm-tools agents rag embeddings prompt-engineering

Agent Platforms Are Four Problems, Not One

You'll deploy a coding agent and think you're done. You won't be told you also need sandboxing, governance, and orchestration — or that only one of these layers is actually mature.

Graham Rowe · April 02, 2026 · agents mcp ai-coding

We Audited crewAI's AI Dependencies: Here's What the Data Says

47 dependencies, 8 scored against 220K repos. World-class choices across the board — and the one category where the landscape hasn't settled yet.

Graham Rowe · April 02, 2026 · agents vector-db rag llm-tools transformers

The Claude Code Ecosystem: Everything You Can Plug In

2,400 repos. 370 new ones per week. A practitioner's guide to what's mature, what's emerging, and what's noise.

Graham Rowe · April 02, 2026 · agents mcp ai-coding llm-tools vector-db

Agent Memory in 2026: What Actually Works for Persistent AI

977 repos, 5 domains, 10+ names for the same concept. A decision guide for builders navigating the most fragmented category in AI infrastructure.

Graham Rowe · April 01, 2026 · rag agents vector-db mcp embeddings

Giving AI Agents Eyes: Browser Automation in 2026

Data extraction, agentic browser control, and stealth infrastructure — scored on quality daily. A decision guide for builders navigating scraping, automation, and the anti-detection arms race.

Graham Rowe · April 01, 2026 · perception agents mcp

How Agents Acquire Skills: The Emerging Architecture of Composable Capabilities

Skills, registries, marketplaces, and the trust problem — scored on quality daily. A map of the 2,700+ repos building the capability layer for autonomous agents.

Graham Rowe · April 01, 2026 · agents llm-tools

AI Agents for Obsidian and Personal Knowledge: What Actually Works in 2026

From established plugins to MCP bridges to experimental agent tools — scored on quality daily. A decision guide for the Obsidian user who wants AI that respects their vault.

Graham Rowe · April 01, 2026 · agents llm-tools rag embeddings

Agent Governance in 2026: Who's Building the Guardrails?

Sandboxing, policy enforcement, security scanning, and compliance — scored on quality daily. A decision guide for teams deploying autonomous agents who need governance tooling that actually works.

Graham Rowe · April 01, 2026 · agents llm-tools mlops

Embeddings Are Easier Than Whatever You're Doing Instead

You're writing regex, tuning Elasticsearch, or building keyword indexes. Meanwhile, 20 lines of code and a $0.02 API call would give you better results. Here's how to stop overcomplicating it.

Graham Rowe · April 01, 2026 · embeddings vector-db rag

Things AI Won't Tell You About Building a Voice App

You'll ask for a TTS library and get one. You won't be told you also need STT, intent handling, orchestration, and evaluation. Here's the full stack, with opinionated recommendations at every layer.

Graham Rowe · April 01, 2026 · voice-ai agents embeddings nlp

Choosing a Voice AI Library in 2026: What's Actually Worth Building On

TTS, speech recognition, and voice agents — scored on quality daily. A decision guide based on which projects are thriving, which are dying, and what the trends mean for your stack.

Graham Rowe · March 31, 2026 · voice-ai agents nlp

How OpenClaw Went from Launch to 1,299 Repos in 8 Weeks

The anatomy of an AI ecosystem forming in real time — what gets built first, what comes next, and where the real adoption signals are.

Graham Rowe · March 30, 2026 · agents llm-tools rag mcp voice-ai