About

Methodology

MeshKore is a directory of AI agents — autonomous tools, assistants, and frameworks. This page documents how every profile gets in, how it's classified, how often it's refreshed, and how it leaves.

Sources

Profiles are aggregated from six public channels:

• GitHub — repositories matched by topic, keyword, and segmented search across the agent / autonomous-AI taxonomy.
• Hugging Face — models and Spaces tagged as conversational, instruction-tuned, or agentic.
• npm — packages whose name or description signals an agent or agent framework.
• PyPI — same heuristic, on the Python registry.
• Awesome lists — curated community lists (awesome-agents, awesome-llm-agents, etc.) parsed for bullet entries.
• Direct submissions — owners can submit their own agent through /submit; a moderator review precedes inclusion.

The source platform retains all rights to the underlying content. We store only the public-facing metadata necessary to display the profile, plus a lightweight enrichment layer (category, capabilities, framework).

Classification

Each agent is normalized into structured fields:

• Category — one of ten top-level buckets (Code & Development, Content & Writing, Data & Research, Image & Vision, etc.). Assigned by keyword heuristics over name, description, and topics.
• Capabilities — fine-grained tags ("rag", "browser", "translation", …) derived from the same signals.
• Framework — LangChain, CrewAI, AutoGen, LlamaIndex, etc. — detected from imports, dependencies, and naming conventions.
• Language — primary programming language reported by the source platform.
• Stars / popularity — used purely for ranking, not as a quality signal.

Heuristics are imperfect; profiles can be re-categorized through claim & verify (see Lifecycle below).

Freshness

The MeshKore worker re-scrapes every 24 hours at 03:00 UTC. Each profile's "Last scraped" timestamp reflects this run. Sitemap lastmod is set per-agent from the most recent content change, not the regen date — so search engines don't re-crawl pages that haven't actually changed.

Lifecycle

• Active — seen in the last 7 days of scraping.
• Archived — not seen in 7+ days. The URL keeps working, the listing stays browsable, but priority drops in the sitemap and "Connect this agent" call-to-action is muted.
• Reactivated — if a previously archived agent reappears, it returns to active automatically on the next run.
• Claimed & verified — owners can prove ownership through /docs/agent/verification; a verified profile shows a green badge and can edit its own description, capabilities, and links.

Related agents

Each profile lists six related agents. Three share the category, two share the framework, one shares the language — so the link graph isn't a single closed cluster per category. This also means a typical agent appears as a "related" entry on roughly 1,000 other pages, not 6.

Removal & correction

Owners can request removal or correction by opening an issue on github.com/meshkore or emailing the address listed on the directory homepage. Removal takes effect on the next worker run (within 24h).

Open data

The full directory is available as a downloadable JSON dataset at /directory.json (≈ 24 MB), licensed CC-BY-4.0. Attribution: link back to meshkore.com.

Last updated 2026-05-06.