A: Hi,

I completely understand the thoughts you are echoing. I would like to explain our approach and why it's actually harder than it seems.

We use credits instead of tokens because our AI agents are autonomous. When you ask the Lead research agent to research about a prospect and prepare a report, I honestly dont know if it will do 5 searches or 15 to get you quality results. We do guide the model with budgets but it's not a fixed workflow. We optimize for outcomes, not token usage.

Here is what makes it complex:

- Different tasks use different models ( larger flagship models for agent orchestration, smaller models for summaries ). Even among these some are good at instruction following, some at coding and some at writing. So I may need Claude to code my emails but OpenAI models to write them. We use image generation models and we may have voice AI as well as video AI as we increase our ambitions and add new features. Possibilities are emerging and we cannot have a fixed roadmap.

- Agents use tools like web search, data enrichment, costs of which have to be factored in, not just tokens.

- Input & Output varies wildly - a workflow might generate 10 lines of config with 10K lines of instructions vs 50 lines of email content.

Other platforms handle this differently:

- ChatGPT/Claude just say "usage will be limited" for $20/mo without specifics
- Bolt/v0 give you X million tokens but don't differentiate between models.
- Some charge per model but then you're optimizing for cost not results.

Token-based pricing sounds transparent but gets messy fast.
- Do you charge differently for input vs output tokens? What about tool usage?
- BYOK (bring your own key) seems good until you hit rate limits or get surprised by bills from autonomous agents making decisions.

Additional credits pricing is something that we provide based on individual requirements. Because tier 4 & 5 have sufficient credits for high usage. Thats why we ask people to contact us, sometimes we just reset credits for reasonable usage without any charges. At the end of the day we want to encourage more use.

We chose credits to abstract this complexity as things change rampantly in the AI space.

Open to suggestions on how to make this clearer.

regards,
Swami