Why We Stopped Using ChatGPT (And Built Our Own AI Workspace)

We didn't stop using ChatGPT because it got worse. It got better.

We stopped because of what our browser looked like at 4pm on a Tuesday.

ChatGPT in one tab. Claude in another. Gemini open because it's better at the long doc. A scratch file where we pasted Claude's answer so we could ask GPT to poke holes in it. A Slack thread where a teammate asked, “Wait, what did the model actually say?” and the honest answer was: it's in my history. You can't see it.

That's not a tooling problem you fix with a better prompt. That's the shape of the tool being wrong.

So we built flAim.

The three things that actually broke

Every AI chat app on the market is built on the same three assumptions. All three stopped being true for us.

1. 1

   You want one model.You don't. You want the fast, efficient one for the first pass, the careful one for reasoning, and the long-context one for the 40-page PDF. Picking a single provider means picking a single set of weaknesses and living with them all day.

2. 2

   You'll do the routing yourself. This is the copy-paste tax. Ask model A. Copy. Paste into model B with “critique this.” Copy. Paste into model C with “now write the final version.” You are the orchestration layer: a very slow, very expensive orchestration layer made of meat and command-C.

3. 3

   AI work is solo work.Chat history is private by default. So the most useful artifact your team produced this week, the thread where someone actually figured it out, is locked inside one person's sidebar. You share the screenshot. Never the thread.

Fix those three and you don't have a better chatbot. You have a different category of thing.

Pillar 1: Use any agent, in one workspace

In flAim, a workspace is a room. You add agents to it. An agent is a chat backed by a model you picked. OpenAI, Anthropic, Gemini, DeepSeek, xAI. Add as many as the work needs.

They live in one shared thread with one compose bar, the Smart Bar. Type @ to send to one agent. Send without a mention and it can fan out to several at once.

The first time you ask three models the same question and watch the answers land side by side, you notice something uncomfortable: they disagree more than you assumed. That disagreement is signal. You were just never in a position to see it before, because seeing it used to cost you three tabs and four minutes.

No lock-in to one lab's roadmap. When a better model ships, you add it. That's the whole migration.

Pillar 2: One prompt, and flAim does the multi-model work

This is the part that killed the copy-paste tax for us, and it's the part we're proudest of.

Switch on Relay and flAim stops being just a chat app and starts conducting the work. You give it one prompt. It:

1. Writes an execution plan: which steps, which agents, and what order. The plan is visible in your workspace, not hidden.
2. Runs the steps, handing work between agents so each can plan, draft, critique, or refine.
3. Pulls the work together into one stronger result, with the thread and context still there for you to inspect.

So “draft the migration guide for our v2 API, check it against the current spec, and flag anything that'll break existing clients” is one send. Not eleven.

Pillar 3: Share the workspace, not the screenshot

Every AI workspace should have been multiplayer from day one. Ours is.

Share a workspace. Invite a teammate. They see the agents, the thread, and the shared context. Live drafts sync while you're typing. Presence shows who's in the room.

The behavior change is bigger than it sounds. Instead of “here's what Claude told me,” it becomes “jump in the workspace, ask it yourself, the context is already loaded.” Prompting stops being a private craft skill and becomes something a team does together, in the open, where the good prompts are reusable and the bad ones get fixed.

A heads-up on billing: when a collaborator runs a managed chat in a workspace you own, it spends your credits. The person who opened the room is buying the round. In a Corporate Team account, that spend comes from the shared team wallet instead.

Starting costs nothing and doesn't need an API key

We made a deliberate call: no key wall. Every plan includes prepaid credits, and top-up packs price cleanly at 100 credits = $1, so topping up never requires a token calculator.

Monthly included credits don't roll over. Purchased credit packs ($5 for 500, $10 for 1,000, or $25 for 2,500) last until you spend them and are available on Pro and Power. If you'd rather use your own provider account, BYOK is optional on paid personal plans, and those prompts don't touch flAim credits. It's an option, not a setup step.

Should you actually stop using ChatGPT?

If you send one question at a time, to one model, alone: genuinely, no. ChatGPT is great at that and you don't need us.

But if your workflow has turned into a tab-management exercise, if you're the one shuttling text between models, or if your team keeps re-deriving the same context in five separate private chats, that's the shape of problem we built for. That was our 4pm on a Tuesday.

We're building this in the open and we read everything. If you try it and something's clumsy, tell us. We'd rather hear it from you than not hear it.