local AI agent workflow

A local AI agent workflow is one where the project records, the tooling that coordinates agents and the runs themselves live on your own computer, instead of in a hosted dashboard you log in to. The appeal is concrete. Your code and notes do not need to be mirrored into someone else's database to be organized, there is no extra account to manage, and the tool keeps working while you are on a plane. The cost is equally concrete. You are the operator. You install it, you keep it running, you back it up, and you accept that it works on the operating systems it supports and nowhere else. It is a trade, not a virtue. It suits people with sensitive repositories, people who dislike another subscription surface, and solo developers who would rather own their records. This page lays out what local means in practice and where Liquid Brain sits.

What local does and does not mean

Local describes where the project record and the coordinating software live. It does not mean the whole workflow is offline. The agent itself, whether a Claude or Codex CLI, usually talks to a model provider over the network, and that traffic is governed by that tool's own settings, not by your project manager. So a local workflow reduces how many places your project records exist, but it does not remove the model provider from the picture. Be precise about this when you evaluate a tool, and ask two separate questions: where do my cards, history and evidence live, and where does my code go when an agent works on it. The first is decided by the project tool, the second by the agent you chose. Mixing them up leads to either false reassurance or needless suspicion.

What you take on in return

Running things yourself has real overhead. Updates are your job. Backups are your job: if the workspace sits on one disk and the disk fails, so does your record. Anything scheduled depends on your computer being on and the application running, which is a different reliability profile from a hosted service that runs while you sleep. Collaboration is harder, because there is no shared server by default, so inviting a second person needs a deliberate mechanism and trust decisions you have to understand. Platform support also narrows to whatever the software was built for. None of this is a reason to avoid local tools. It is the reason to choose with open eyes, and to check that backup and restore are first class features, not an afterthought you discover during a failure.

Where Liquid Brain sits

Liquid Brain is a Windows desktop application, the Desk, and its current source candidate requires CPython 3.14 AMD64 and pinned dependencies [C10]. The compatibility statement is Windows only, with no promise of other platforms or arbitrary CLI compatibility [C10]. You can open Claude or Codex from a card, which needs a configured, compatible CLI [C1b]. Scheduled tasks work only while the Desk is running [C1c]. You can export a workspace backup and restore it into a new location with dispatch paused, so you review roots, credentials and readiness before agents resume [N05]. Imported project work can be previewed and mapped, with conflict aware apply [N04]. Backups do not resume agents by themselves, and roots, credentials and readiness need review after a restore [N05]. Project scoped collaboration and invitations exist in the candidate but remain under validation [C8].

QUESTIONS

Does a local AI agent workflow mean my code never leaves my computer?

No. The project tool being local describes where cards, history and evidence live. The agent CLI you run, Claude or Codex, communicates with its model provider under its own settings. Check that tool's data handling separately from the project manager's, because they answer different questions about where your code goes.

Do I need an account to use Liquid Brain?

The approved facts describe a Windows desktop Desk that runs on your computer [C10] and do not describe an account or hosted service. Your Claude or Codex CLI has its own sign in. No purchase option exists yet, since the product is marked COMING SOON.

What happens to scheduled agent tasks if I close the app?

They do not run. Liquid Brain scheduling requires the Desk to be running and is not a background operating system scheduler [C1c]. Plan for that if you expect overnight work, keep the application open, and review what was missed when you reopen it, since a missed task is not the same as a completed one.

RELATED GUIDES

• What is a project management harness?
• Cards vs scheduled tasks: what starts an agent?
• How to preview work with the Liquid Brain harness CLI

RELATED HARNESS PAGES

• AI agent project memory AI agent project memory explained: why durable project records beat chat history, and how Liquid Brain keeps context with the work.
• scheduled agent tasks Scheduled agent tasks done carefully: explicit prompts, readiness and authority checks, and why a started run is not a finished one.
• multi-agent project coordination Multi-agent project coordination: keep Claude and Codex work on one project organized with shared cards, scope, authority and review.