Agents
A map of the agent landscape: self-hosted personal frameworks, hosted teammate products, coding CLIs, and libraries for building your own. Links go to the official repo or product page we could verify.
Checked as of Sep 3, 2026. Star counts, maintainers, and even repo locations move fast — especially in the OpenClaw family — so confirm the URL before you install anything that asks for system access.
1. Open-source personal agent frameworks
Self-hosted daily-driver assistants — gateway runtimes, skills, and persistent memory you run yourself.
The original gateway platform; TypeScript, with a large skill ecosystem.
Nous Research — self-improving agent that creates its own skills, with persistent memory.
HKUDS — Python rewrite, positioned as auditable in an afternoon.
Lightweight TypeScript alternative (~hundreds of lines of core); each agent runs in its own locked container.
Canonical repo is nanocoai/nanoclaw (qwibitai/nanoclaw redirects here).
Rust runtime aimed at tiny hardware footprints (very low RAM).
Sipeed — Go, single static binary, targets cheap boards.
Also in this family, without a single verified product URL in our check: TrustClaw, Goclaw, IronClaw, NullClaw, OpenAGI. Names circulate; treat installs as untrusted until you confirm the official source.
2. Hosted AI teammate products
Productized assistants — mostly cloud / subscription, sometimes with a local or open-source option.
xAI — named bots on a persistent cloud computer with browser, filesystem, terminal, connectors, and durable state.
Anthropic — agentic knowledge-work mode across desktop, web, and mobile; multi-step tasks on files and tools you choose.
Cloud agent for long-horizon autonomous tasks.
Cloud agent orchestrator that coordinates multiple models to run multi-step workflows in the background.
Personal assistant as a native Mac app or Vellum Cloud, with iOS, web, voice, email, Telegram, and Slack surfaces sharing one memory.
Open-source core (MIT) with managed cloud and self-host options — not a closed-only product.
3. Agentic coding CLIs
Agents scoped to a codebase — terminal-native plan / edit / test loops.
Anthropic — terminal coding agent.
OpenAI — open-source coding agent CLI.
xAI — terminal-native coding agent with parallel subagents and MCP support.
Google — open-source Gemini coding agent in the terminal.
4. Developer frameworks
Libraries and automation platforms for building your own agents.
Stateful, graph-based agent workflows (LangChain ecosystem).
Role-based multi-agent crews.
Microsoft — multi-agent conversation framework.
TypeScript agent framework.
Workflow automation with agent nodes.
Anthropic — the harness behind Claude Code, exposed as a Python / TypeScript library.
How to read this
Category 1 projects launched in a tight cluster and still churn. Category 2 mixes closed cloud products with hybrids like Vellum (open-source core). Coding CLIs and frameworks are adjacent but not the same job as a daily-driver personal agent — we keep them separate so the map stays useful.
When we publish a Hands On guide for an agent here, it shows under that tile. Browse all guides on Hands On.
