All Blog

Stop copying between AI windows: how local multi-bot division of labor works

Separate ChatGPT, Claude, and coding tabs mean endless paste and muddy roles. Local multi-bot @ handoff keeps each job's memory and spend apart.

Stop copying between AI windows: how local multi-bot division of labor works

Three tabs open: ChatGPT for draft, Claude for rewrite, a coding agent for the repo. You paste the brief into the first, then the second, then a third time into the coding window because the first two never saw the file path. Halfway through, nobody remembers which version of the ask is current. The problem is not that any single model is dumb. It is that tabs are not a team.

Why browser tabs fail as a "team"

A tab is a private chat with one assistant. Context lives inside that window. When you need another model or another role, you open another tab and re-explain. Copy-paste becomes the handoff protocol. Roles blur: the same window that scouted sources starts rewriting the landing page, then answers a support-ish question because you did not want to open a fourth tab. By evening the thread is mush, and you cannot tell which paste introduced the wrong constraint.

Spend is just as opaque. Each product has its own meter or weekly pool. Cross-window work does not show "research burned this, drafting burned that" in one place. Public chatter around multi-bot products often ends the same way: after several bots run for a while, the bottleneck is quota and shared group context, not a missing model (see people leaving cloud seats for local multi-bot shells). Some people leave a cloud multi-bot seat when the weekly allowance is gone. That is a usage pattern, not a ranking of brands.

The other failure mode is one overloaded worker. Pile enough skills and MCPs onto a single executor and writing, retrieval, and monitors share one muddying context while the bill climbs; the OpenClaw news-monitor story in the path post is exactly that. It looks different from three messy tabs, but the fix is the same idea: clear roles and a handoff that does not depend on you pasting the brief again.

What division of labor looks like on a local multi-bot desktop

A local desktop multi-bot setup (Teebots is one example) puts named bots in one group chat on your machine. You @ the research bot, then @ the drafting bot; they can also pass work to each other. Each bot keeps its own personal memory and tool scope. Other bots do not read that private memory. There is no shared group memory that mixes every role into one pile.

Keys stay on this computer (BYOK). The app is free; model spend lands on your provider bill, so matching which bot burned tokens is clearer than staring at three product UIs. Bots can use the browser; scheduled tasks can run. Sensitive steps such as submitting a form, paying, or entering a password stop and ask you first.

This is not "full compatibility with every cloud multi-bot preset," and it is not a marketplace one-click install story. It is a desk where roles, memory boundaries, and the provider bill sit with you. For path contrast (local team vs one cloud assistant), see A local team of bots, or one cloud assistant?. For weekly quota burn, see Cloud quota burned out? Move to a local BYOK multi-bot setup. For why keys should not sit in a hosted console, see Why not hand your API keys to a host.

Role splits that stay practical

Keep the roster short. A one-person company might run three bots: one watches public pages and notes, one turns decisions into a checklist, one drafts the outward note. A creator might split scout, draft, and competitor watch. Name the job, not a costume. The point of @ handoff is that research memory stays in research, and drafting does not inherit yesterday's half-wrong paste from another window.

If cross-window handoff keeps losing state unless you write it down yourself, that is a signal you need persistence per role, not another tab (see handoff only counts once it is written down).

Who should not bother

If you need the same always-on cloud PC on every device, a local desk will not replace that. Stay on the cloud path, or compare dimensions (where it runs, how models are chosen, whether you pay another seat) in Teebots vs Grok Bot. Local multi-bot helps when the files, browser, and logins already live on this machine, and the pain is paste hell plus muddy roles.

How to start

  1. Open teebots.ai/download and install the build for your system.
  2. Add your own model key (BYOK).
  3. Create two or three short role briefs. Not ten bots on day one.
  4. Open one group. Assign work with @. Watch the provider bill for who burns tokens.

Stop when copy-paste between AI windows is the main workflow. Division of labor is @ handoff plus personal memory boundaries, not a longer feature checklist. Quota migration is covered in Cloud quota burned out?; key custody in Why not hand your API keys to a host.

Your local team of bots

Not locked to one model vendor. Each bot uses your own key — Claude, OpenAI, DeepSeek, Kimi, or any compatible endpoint. Requests go straight to the provider.