Skip to content

How Bazilion is different

Bazilion is openly OpenClaw-inspired: it borrows OpenClaw’s prompt-only skill format, and it can import skills straight from a local OpenClaw install. But it is a different kind of tool with a different center of gravity. If you know OpenClaw, this page explains what’s the same, what’s different, and why.

OpenClaw is a personal AI gateway — it connects one assistant to the many channels you already use (WhatsApp, Telegram, iMessage, Signal, and more) and extends it with a rich plugin system.

Bazilion is a multi-agent runtime — it’s built around spawning teams of agents that share a workspace, a memory, and a mailbox, and coordinate on work. The unit isn’t “one assistant on many channels”; it’s “many agents collaborating in one place.”

A profile is an agent template; a profile group is a team template. One action spawns an ordered set of agents — each with its own name, model, and reasoning level — into a shared group. This team-first model has no direct OpenClaw equivalent. See Core concepts.

Every agent belongs to exactly one group, which provides one filesystem root (optionally a symlink to a real project), one USER.md of shared context, one mailbox, and one memory. Agents work on the same material rather than each living in a private silo.

Group memory is markdown indexed with qmd (BM25 search). Any agent in the group can write a decision once and another can find it later by searching — memory is a team resource, not a per-agent log.

Agents coordinate through durable messages (send_message, read_inbox, wait_for_reply), not just live chat. Combined with triggers (interval and cron wake-ups), an agent can pick up work left by another and act on a schedule.

OpenClawBazilion
ShapePersonal assistant gatewayMulti-agent team runtime
ExtensibilitySkills and plugins (in-process TypeScript, ~28 lifecycle hooks)Prompt-only skills and out-of-process MCP servers; all other capability is native daemon code
ChannelsMany first-class transports (WhatsApp, iMessage, Signal, IRC, …)HTTP API + web UI + CLI, plus Telegram
Config & secretsopenclaw.json (JSON5) + credential filesA single SQLite database (encrypted secrets, plaintext config)
StateWorkspace + config filesOne daemon owns ~/.bazilion; agents, groups, sessions, and memory are inspectable files
RegistryClawHub (thousands of community skills)None — import skills from a directory, zip, or a local OpenClaw

Bazilion runs a single daemon that owns the database, the scheduler, secrets, and the agent runtime. The CLI, web UI, and any future client are stateless and talk to it over HTTP. Each chat turn even runs in its own short-lived worker subprocess, so one turn can’t corrupt another. This keeps the security and state model small and auditable.

Bazilion is a personal, MIT-licensed project. It runs on your machine, with your provider keys, storing state under ~/.bazilion. There’s no hosted service and no account — the daemon binds to localhost by default, and exposing it to your LAN or a phone is an explicit opt-in.

  • No plugin SDK. OpenClaw’s plugin system (new channels, providers, tools, and lifecycle hooks via in-process code) is powerful but is also arbitrary code running inside the gateway. Bazilion keeps extensibility to prompt-only skills, out-of-process MCP servers, and native code, so a third-party extension can’t crash the runtime or intercept every tool call.
  • Fewer channels. Bazilion isn’t trying to be everywhere you chat. It exposes a clean HTTP API (with a web UI and CLI on top) and a single, deeply-integrated Telegram surface.
  • No registry. Skills come from a local directory, a zip, or your existing OpenClaw install — there’s no marketplace to browse.

Reach for OpenClaw if you want one assistant reachable across every messaging channel, with a large catalog of plugins and community skills.

Reach for Bazilion if you want to design small teams of agents that share a project, a memory, and a way to talk to each other — and you want all of that running locally, under your control, as inspectable files.