Entities

Anthropic's hosted agent infrastructure — production-grade agents run on Anthropic's platform rather than self-hosted. Entered public beta in April 2026 with Notion, Rakuten, and Asana as launch customers.

15 stories

AI safety company and maker of the Claude family of models. Operator of Claude Managed Agents and a central player in the agentic-AI infrastructure layer ClawBlog covers.

10 stories

The most-starred open-source AI agent framework in GitHub history — a general-purpose autonomous agent that operates over everyday messaging platforms (Discord, WhatsApp, Telegram) through a library of 100+ AgentSkills.

10 stories

Anthropic’s coding-focused agent harness — a command-line tool (also available in desktop, web, and IDE forms) that wraps Claude in an agent loop with file, shell, and tool access to do real software engineering work.

8 stories

Nous Research's self-improving agent with persistent memory — an MIT-licensed framework that runs across five execution backends (local, Docker, SSH, Singularity, Modal), and one of the fastest-growing agent frameworks of 2026.

6 stories

AI research company and maker of the GPT family. Relevant to the agent ecosystem as a frontier-model provider and as the company OpenClaw founder Steinberger joined in February 2026.

5 stories

The failure mode where an agent follows instructions hidden in content it fetches, because it cannot reliably tell data from commands. The defining unsolved security problem of agents that act on the open web.

5 stories

The category ClawBlog covers: software agents that act autonomously in the world — planning, calling tools, and executing multi-step tasks — layered above frontier models. The technology, business, use cases, and security of that layer.

3 stories

The public skill registry and marketplace for OpenClaw — a catalog of 13,700+ AgentSkills with a 90/10 revenue split for paid skills. Hardened with VirusTotal scanning after the ClawHavoc supply-chain attack.

3 stories

A widely used, model-agnostic framework for building LLM and agent applications. One of the open frameworks now contending with the move toward leaner, lab-blessed first-party agent SDKs.

3 stories

An open protocol for connecting AI agents to external tools and data sources through a standard interface, so a capability built once can be reused across any MCP-compatible agent instead of re-integrated per framework.

3 stories

The agent capability of operating software through its GUI — reading the screen, moving the mouse, typing — rather than through APIs. The most general agent skill, and the one with the widest risk surface.

2 stories

A packaged unit of capability an agent loads to gain a new action — OpenClaw’s building block, distributed via ClawHub. Convenient and composable, but each skill runs with the agent’s access, which makes provenance a security question.

Microsoft’s open framework for multi-agent systems, organized around agents that converse to solve a task. A recurring comparison point for how to coordinate a swarm.

The early-2026 supply-chain attack on the OpenClaw ecosystem: typosquatted malicious AgentSkills published to ClawHub. The incident drove ClawHub’s VirusTotal scanning partnership and a broader reckoning over agent-skill provenance.

A multi-agent orchestration framework that organizes agents into roles and "crews" that collaborate on a task — one of the recurring comparison points for how to coordinate a swarm.

A hosted business-agent product from RunDemon: an AI teammate with its own machine, browser, inbox/channels, memory, and activity log, aimed at small businesses and agencies.

Google’s Agent Development Kit — a Gemini-native framework for building, composing, and operating agent applications.

Google's AI research division and maker of the Gemini family and the Nano Banana image models, including the high-fidelity Nano Banana Pro used in agent image pipelines.

Hostinger’s one-click VPS application for Hermes Agent, emphasizing a browser admin panel, multi-platform messaging, 200+ model routes, scheduler, and self-hosted key/data control.

Hostinger’s one-click VPS application for deploying OpenClaw with Docker manager, logs, updates, and AI-managed VPS infrastructure.

Hostinger’s one-click VPS application for Paperclip, emphasizing self-hosted orchestration, role hierarchies, per-agent budgets, audit trails, and BYO agent/model runtime.

An open protocol for connecting AI agents to external tools and data sources through a standard interface, so a capability built once can be reused across any MCP-compatible agent instead of re-integrated per framework.

Google DeepMind's high-fidelity image-generation model (April 2026) — the premium tier of the Nano Banana line, used in agent pipelines when output quality matters more than cost.

Open AI research collective and maker of Hermes-Agent, known for permissively licensed models and agent tooling.

Ollama’s OpenClaw integration path: launch OpenClaw through Ollama, connect local or cloud models, and use recommended large-context models for agentic tasks.

OpenAI’s first-party agent application SDKs for building tool-using, guardrailed agent workflows in Python and JavaScript.

A model-routing service that puts one API in front of many providers’ models, with per-call usage and cost accounting — a common way agent builders compare models and meter spend across the ecosystem.

A multi-agent orchestration layer for "zero-human companies" — a Node.js orchestrator and React control UI that coordinates fleets of agents via a heartbeat protocol, budget enforcement, and a shared ticket queue.

A pattern where an agent retrieves relevant external information at run time and feeds it into the model, grounding output in real data rather than relying on what the model memorized. A core primitive of agent memory and accuracy.

A long-running malware-scanning service that aggregates many antivirus engines and threat signals. After ClawHavoc, ClawHub adopted it to scan submitted AgentSkills for known-malicious payloads before listing.