Hoplite Cloud Coding Agents: Deploy Your Whole Setup in the Cloud
The onboarding story is the strongest part. When you sign up, Hoplite ports your local setup — sessions, memories, MCP servers — and gets your project ready to run in the cloud. That's a deeper integration than most cloud agents offer. Typically you'd recreate your environment from scratch; Hoplite treats your existing local config as the source of truth. For anyone running multiple MCP tools or long-lived memory sessions, this could be the thing that makes cloud agents actually usable. It also means the learning curve is gentler because your agent literally remembers what it knew locally.
On the agent side, they made a deliberate bet: custom harness. It gives them independence from Anthropic and OpenAI release cycles, and they can test experimental features before the model vendors catch up. The tradeoff is clear — they don't inherit the battle-tested tooling of Codex or Claude Code out of the box. But for a startup focused on agent workflows, ownership seems worth the risk. They even let you connect your existing Codex subscription and route OpenAI models through it, which softens lock-in concerns.
The infrastructure stack is worth noting: AWS for hosting, Temporal for durable workflows, Modal for sandboxes, PlanetScale for the database. The Temporal pick stands out because long-running agent tasks are exactly what durable execution was designed for. If an agent gets interrupted mid-task, that workflow engine can resume cleanly rather than losing state. That's the kind of reliability you want when running hundreds of agents concurrently.
The bigger vision: as models improve, developers will review less code and instead review product output — visually checking new flows, verifying APIs, testing CLIs on different OSes. Hoplite wants to be the platform where you run hundreds of agents and QA whole features by watching the results.