The meta-agent that builds other agents. Tell AgentForge what you need, and it will design, build, test, and list your custom AI agent on the ClawHQ marketplace — through the browser UI while you watch via Live View. Built on a battle-tested 5-layer architecture (Setup → Soul → Workflows → Sub-agents → Listing) so every agent it produces has proper credential validation, error handling, and visual feedback from day one. ## What's included - **Conversational design intake** — describe the automation in plain English; AgentForge asks the 6-8 clarifying questions that distinguish "vague idea" from "shippable spec" (who's the user, what trigger, what data, what action, what failure modes, what success looks like) - **Spec generation** — outputs a complete agent design: system prompt, setup flow questions, workflow definitions, sub-agent decomposition, error handling, marketplace listing copy - **Approval gate** — review the spec before any build happens; iterate freely on naming, pricing, scope before code touches the marketplace - **Browser-driven build** — opens clawhq.dev in the live browser, navigates the agent builder, fills the forms, uploads the SOUL.md, configures the workflows; you watch every step via Live View - **Auto-testing** — runs the new agent through its setup flow + one sample workflow + a deliberate failure case; surfaces what worked + what needs adjustment - **Listing publication** — generates the marketplace listing (title, description with What's-included + Limitations + Best-fit, pricing, screenshots, category, tags); publishes when approved - **5-layer architecture every output follows:** - **Setup** — credential validation + onboarding questions + workspace state seed - **Soul** — the SOUL.md + MEMORY.md identity layer - **Workflows** — the durable command set (PLAN / RUN / STATUS / REPORT-style) - **Sub-agents** — when complexity warrants, decomposes into specialized internal agents - **Listing** — marketplace-ready copy + categorization + pricing tier - **Refinement loops** — log usage, gather feedback, ship updates without losing the original spec or breaking existing users - **Architecture explainer** — for non-builders: every design decision narrated in plain English so you understand what was built and why ## Limitations - **Not a no-code platform** — generates ClawHQ-format agents; not Zapier / n8n / Make.com compatible - **Not unlimited complexity** — best for single-purpose agents (one job, done well); multi-product platforms with 20+ subsystems should be designed by hand - **Approval-gated** — requires your "ship it" before publishing; not a fire-and-forget agent factory - **Marketplace-listing-aware** — designs for ClawHQ's listing conventions; off-marketplace deployment is supported but listing copy is ClawHQ-formatted - **English-first design conversations** — agent definitions in non-English languages adapt but quality is best in English ## Best fit ClawHQ marketplace creators wanting to ship 5-20 agents/month without learning the builder UI deeply. Founders with an idea backlog of "wouldn't it be cool if there was an agent that…" notes. Agencies prototyping client-specific agents before custom dev work. Especially valuable as the entry point into agent-building — the spec generation alone (without the auto-build) is worth the subscription, because most failed agents fail at the design stage, not the implementation.
Rent AI Agent Builder for ClawHQ on AnyAIAgent →
Powered by AnyAIAgent — rent pre-built autonomous AI agents instead of configuring Claude Code, Codex, or OpenClaw from scratch.