Team bot folders do not isolate workloads
Give a Rakazo Team bot a folder and it can still reach the rest of the Team workspace. The runtime documentation makes that boundary explicit. Per-bot starting directories and separate browser profiles organize work, but Team bots share the computer's OS user and access. Use separate computers when workloads require isolation.
Rakazo supplies persistent bots with conversations, memory, routines, and browser or terminal access. You choose the model and computer provider, then manage the service through web, Electron, or mobile clients. Consider a trial when a bot needs to resume work with persistent files and browser state. For occasional chat, weigh that benefit against operating a database and computer service. The README labels the project beta.
What happened when we ran it
Our sandbox installed Rakazo commit e55263f in 42 seconds and built it in 21 seconds on 2026-10-17. The pnpm run recorded 1,862 dependencies and 1,841 MB on disk.
Tests failed with exit 1 after 332 seconds. Vitest reported 8,605 passed, 26 failed, and 256 skipped out of 8,887. The file summary showed two failed files, 660 passed, and 44 skipped out of 706. The supplied tail lists skipped PostgreSQL and Docker-related cases, then the failed-test totals. It does not identify the underlying exceptions. Attributing the failures to a database, missing package, or product defect would go beyond that evidence.
The sandbox recorded 3 CPUs, 8 GB RAM, and image label lab-node:22. Actual runtime and operating-system versions require raw toolchain logs. Neither the build nor partial suite establishes a successful live-model task. The supplied documentation describes the project separately; its extracts are not pinned to our measured commit.
Published images and source development are separate routes
The README documents a published-image installer requiring Docker Engine, the Compose plugin, curl, and OpenSSL, without a source checkout or Node installation. The source route instead specifies Node.js, pnpm 9, Docker, PostgreSQL configuration, and independent secrets. Our measurements cover the source run. They provide no timing or success result for the image installer.
The source instructions require matching PostgreSQL credentials and independent authentication, encryption, screen-proxy, and sandbox-supervisor secrets. You must also connect a model provider and configure any remote computer provider you choose.
Managed integration catalogs are optional. The README says users can add an HTTPS MCP server, Treg endpoint, or OpenAPI JSON document without enabling those catalogs. For an adoption trial, use disposable accounts and tasks whose results you can inspect.
Desktop and mobile applications are clients of the same backend. The README says bots stay on when the backend runs on a server. Choose where that backend and its storage will remain available before choosing a client. Our run did not measure unattended operation or calls to a model provider.
Persistent workspaces still need independent backups
The runtime documentation describes a local store that keeps the latest workspace copy, rather than an immutable revision archive. Replacement machines import the latest stored workspace, which includes browser profiles. Latest-only storage does not provide a history from which to recover an unwanted edit or deletion. The documentation requires a persistent volume, encryption at rest, and off-host backups for production deployments.
The portable workspace is also different from a disk snapshot. System packages installed outside it are lost when moving to another provider. Keep durable customization in a reproducible image or setup recipe.
For a trial, save a disposable file, replace the computer through the intended lifecycle, and check what returns. Then check whether your independent backup can recover an earlier copy after an unwanted edit. These are recommended checks, not recovery results from our sandbox.
A separate Docker failure report warrants a recovery check
Repository metadata records a push on October 19, 2026, and issue 1527 was opened that day. Those facts do not establish reliability or change the result of our dated run.
Issue 1527 reports accumulated viewer processes, memory exhaustion, and lost desktop control on a shared Docker Team Computer. The report identifies main at 20a0a8e (#1511), linked to commit 20a0a8eb51a95710ef681221c1b437406c350285, on macOS with Docker Desktop. That revision and environment differ from our sandbox. The report is a reason to exercise recovery on your chosen provider; it does not explain our 26 failed tests.
Alternatives for terminal and messaging workflows
Hermes Agent documents a terminal interface and messaging access through a single gateway process. Consider it when your main requirement is working through a terminal and messaging services.
OpenClaw documents a Gateway that connects its Control UI, CLI, TUI, and messaging channels, with companion apps and nodes adding device capabilities. Consider it when access through existing chat channels drives your shortlist. These are documented workflow options, not comparative performance findings or claims that either project lacks persistent workspaces.
Trial Rakazo if its persistent computer model justifies the operating burden. Keep unrelated trust domains on separate computers, maintain an independent workspace backup, and verify recovery with disposable data before connecting valuable accounts or files.