WIP. Better-Agent is the successor to better-chat.

Better Agent is a web-native system for creating scoped agents that do durable project thinking, like research, planning, monitoring, with you. Instead of a chat window, you compose a Thinkspace: a bounded environment with one Goal, its own references, memory, skills, and permissions, and a dedicated agent. You set it up, let it rip, and come back to review the output and continue to iterate until you consider the goal to be completed.

The product is architected around the idea that, while agents are cheap and infinitely parallelizable, your human attention and judgement are a scarce, serial resource. So Better Agent paces and batches agent output through a cross-Thinkspace Review Queue rather than encouraging you to spawn and watch more agents burn tokens. It is deliberately not an orchestration dashboard. See CONTEXT.md for the full domain model and PRODUCT.md for design principles.

Better Agent evolves from Better Chat's codebase lineage but replaces the chat-first model with Thinkspaces. The agent runtime is built on Cloudflare's Project Think (@cloudflare/think)—Durable Object-backed agent loops, session storage, recovery, and tool execution—with one durable Agent per Thinkspace (see .agents/docs/adr/0006). The rest of the stack: Cloudflare Workers, Hono + oRPC, Better Auth, Drizzle/D1, TanStack Start, and Alchemy for IaC, organized as a Turborepo monorepo (apps/web, apps/server, packages/*).

Requires Bun.

bun install
bun run dev  # web + server managed via alchemy.run

Other useful scripts: bun run check-types, bun run check / bun run fix (Ultracite lint/format), bun run db:studio:local, and bun run deploy (Alchemy).