Tably Team
“How much is this going to cost me?” is a legitimate question that most tools dodge. Here is the full answer: what you pay, to whom, how much, and how to bring it down.
Who you pay
Tably Team does not resell tokens. You pay for the models directly to each provider with your API key (Anthropic, OpenAI, Google, DeepSeek, xAI, Groq, Moonshot, Mistral, OpenRouter) or with your subscription (Claude Code with Claude, Codex with ChatGPT). Tably shows an estimate before you start and the real cost when you finish, calculated at each provider’s list price.
The three components of the cost
1. The council (the team debate)
Six agents, several turns each, every turn carrying the context of the previous debate. It is the most variable component, because it depends on:
- The models: a council with Claude Opus and full GPT costs tens of times more than one with DeepSeek, Kimi and Llama on Groq.
- The number of turns: the debate mode and the round settings determine it.
- The size of the context: attached documents, researched references, long briefs.
Typical orders of magnitude for a full debate:
| Team | Typical council cost |
|---|---|
| All premium (Opus, full GPT, Gemini Pro) | A few dollars |
| Mixed (one premium model in leadership, the rest budget) | Under a dollar |
| All budget (DeepSeek, Kimi, Llama, Flash) | Cents |
| Local (Ollama) | Zero |
2. The build (the code)
It is the component that weighs most if you pay via API, and the one that can be zero if you use a subscription:
- Claude Code or Codex via subscription: spends no tokens. It consumes part of your plan’s usage limit.
- Via API: a medium-sized app costs between cents (DeepSeek, Kimi, Mistral) and a few dollars (Opus, full GPT). Each round of changes adds up.
3. The extras
- Reference research: the vision model’s tokens, a few cents.
- Prompt engineering and brief: a couple of calls, cents.
- Publishing: Tably does not charge for publishing on its own instance; Cloudflare, Fly, Neon and Supabase have free plans and charge for usage above them.
How to lower the cost without lowering quality
- Build via subscription. It is the biggest lever: the code is the most expensive part via API and it is free with Claude Code.
- Premium only where decisions are made. A Claude or GPT as CTO and CEO, budget models everywhere else. The debate does not lose quality and the cost drops a lot.
- Use the Direct builder for small things. No council, no debate tokens.
- Learn with Groq or Ollama. Experiment with the platform for free and spend when the project is real.
- Ask for changes in batches. Ten changes in one request cost less than ten requests of one change each.
What Tably shows you
- Before you start: an estimate per agent, with a note on the assumptions it uses.
- During: accumulated tokens and cost in the status bar, with a timer per turn and a total.
- When it finishes: the real cost of the council and the build; in the Direct builder, additionally, the card with the estimated savings compared to the team route.
- On the home screen: the month’s API spend and the total, adding up all projects. Subscription builds do not count toward it.
Compared with the alternatives
- A human team: a small app, several thousand dollars and weeks. Tably: minutes or hours, and the cost of the tokens.
- Credit-based app generators: you pay the provider’s credits at their price, without knowing which model runs behind the scenes. Tably: your key, your model, list price.
- A single model in a chat: cheaper per attempt, but more attempts. The debate beforehand avoids building the wrong thing.
Frequently asked questions
Is Tably free? The platform can be used with your own keys and subscriptions. The only costs are what the AI and infrastructure providers charge you.
How much does a typical app cost? With a subscription and a mixed council, under a dollar in tokens. Everything via API with premium models, a few dollars. Without a subscription or an API, with Ollama, zero.
Can I set a limit? Your providers let you set spending limits in their dashboards. Tably shows the accumulated spend so it does not take you by surprise.
In summary
Build via subscription, put premium models only in the seats that make decisions, and use the Direct builder for small things. With that, a complete app costs cents or nothing in tokens, and what you see in Tably is the real cost, not a marketing estimate.
