AgentGrid.io gives each agent its own folder permissions
AgentGrid.io was the number two launch on Product Hunt on Sunday, and it's a cleaner idea than most of what ships in the "workspace for agents" category. The pitch: "The shared drive for humans & agents." Docs, reports, slides, apps and designs live as live artifacts with version history, and both people and agents edit the same copy.
The part worth noticing is the permission model. You connect Claude, Codex or your own agents "with scoped, revocable permissions", and the four rights, read, write, modify and publish, are "set per agent, per folder, not just per team." Publish is a separate right from write, so an agent can draft into a folder all day without being able to push anything public. Everything is stored and diffed like code, which fits how agents already work. The structured data on the site adds a detail the landing copy doesn't: an agent can create a hosted artifact with one unauthenticated POST to the MCP endpoint and get a claim link back, but anonymous artifacts are read-only and expire in 24 hours unless a human claims them. AgentGrid is a product of Anima App, the design-to-code company.
Why it matters: most agent setups today hand over a token that can do everything the human can do, and the Gheist campaign Wiz described this weekend shows what that costs when a token leaks. AgentGrid's shape, small per-agent rights, a separate publish step, and a human claim before anything persists, is the authority layer applied to documents instead of code. There's no pricing or usage data yet and the Product Hunt thread was small, so treat it as a design to copy more than a proven product.
The day's number one launch was in the same family: Moli by Lexmount, a headless browser engine built from scratch for agents that reads structure first (DOM, text, links, forms, network responses) and renders layout and screenshots only on demand, with Chromium kept as the fallback for fidelity. It claims faster startup and less memory than Chrome but publishes no benchmark yet.
AgentGrid: https://agentgrid.io/ Moli: https://browser.lexmount.com/en
← Back to all articles
The part worth noticing is the permission model. You connect Claude, Codex or your own agents "with scoped, revocable permissions", and the four rights, read, write, modify and publish, are "set per agent, per folder, not just per team." Publish is a separate right from write, so an agent can draft into a folder all day without being able to push anything public. Everything is stored and diffed like code, which fits how agents already work. The structured data on the site adds a detail the landing copy doesn't: an agent can create a hosted artifact with one unauthenticated POST to the MCP endpoint and get a claim link back, but anonymous artifacts are read-only and expire in 24 hours unless a human claims them. AgentGrid is a product of Anima App, the design-to-code company.
Why it matters: most agent setups today hand over a token that can do everything the human can do, and the Gheist campaign Wiz described this weekend shows what that costs when a token leaks. AgentGrid's shape, small per-agent rights, a separate publish step, and a human claim before anything persists, is the authority layer applied to documents instead of code. There's no pricing or usage data yet and the Product Hunt thread was small, so treat it as a design to copy more than a proven product.
The day's number one launch was in the same family: Moli by Lexmount, a headless browser engine built from scratch for agents that reads structure first (DOM, text, links, forms, network responses) and renders layout and screenshots only on demand, with Chromium kept as the fallback for fidelity. It claims faster startup and less memory than Chrome but publishes no benchmark yet.
AgentGrid: https://agentgrid.io/ Moli: https://browser.lexmount.com/en
Comments