Clawdi

Clawdi is a coding tool. Clawdi hosts open-source Hermes or OpenClaw agents as always-on Cloud Agents and links local agents like Claude Code and Codex through a free CLI, sharing memories, skills, vaults and app connectors across all of them.

Clawdi is a platform for running AI agents in two ways: as hosted Cloud Agents or as agents already installed on a user's own computer. Cloud Agents run either Hermes or OpenClaw, two open-source agent runtimes, on Clawdi's infrastructure, so there is no server to set up or maintain. The agent stays on after the user logs off and is reached through chat apps such as Telegram, Discord or WhatsApp. The second path is a free, MIT-licensed command line tool. After installing it, signing in with a login command and running a setup command, local agents such as Claude Code, Codex, Hermes, OpenClaw, Pi or OpenCode can be linked to the same account. Feature support varies by agent. What ties both paths together is shared context: memories (durable facts and preferences), skills (reusable instruction files any agent in a project can use), vaults (secrets scoped to specific projects) and connectors that add apps like Gmail, Notion, GitHub, Slack and Linear through OAuth. In practice, the typical use is delegating routine knowledge work by chat message. The site lists email triage, calendar management, report and document summaries, CRM updates, spreadsheet analysis, web research monitoring and coding help as examples. Deployment is a three-step choice of runtime, AI access (Clawdi AI billed pay as you go, or the user's own provider key) and chat channel. Cloud plans are tiered by compute and storage, each hosting one agent. Clawdi positions itself against closed, vendor-hosted assistants by stressing inspectable source code, model choice and control over where data lives. The CLI, backend and dashboard are open source and can be self-hosted. Its scope is individual accounts: each Cloud Agent belongs to one account, and team needs are handled through a support conversation. It suits developers and technical operators who juggle several agents and want one memory layer, more than users who want a fully packaged consumer assistant.