ComputeFlux: The First OpenAI-Compatible AI Gateway for Polkadot

Ecosystem
21

Thank you for sharing this. We have been very focused recently on getting the ComputeFlux beta ready and supporting real users, so we may have missed some of the latest changes regarding ecosystem funding.

If third-party projects are indeed no longer being considered for funding, I would be disappointed, especially for teams building products specifically for the Polkadot ecosystem. That said, our motivation has always been to build something useful. Whether we receive funding should ultimately depend on the community’s assessment of the value we create, and we will continue building regardless.

We are also open to VC funding, but we currently have limited connections to VCs that understand both decentralized AI infrastructure and the Polkadot ecosystem. If you know of any other funding opportunities, programs, or relevant investors we could speak with, we would sincerely appreciate any guidance or introductions.

22

Andreessen Horowitz (a16z)

Paradigm

Sequoia Capital

Pantera Capital

Polychain Capital

Digital Currency Group (DCG)

Blockchain Capital

Electric Capital

Multicoin Capital

Binance Labs

Jump Crypto

Framework Ventures

NGC Ventures

23

Try the beta test for free on computeflux:

24

Stumbled on this thread this week — finally tried the beta for the first time. A few early impressions and questions:

1. **How do the beta credits map to actual cost?** Planning real runs but haven’t seen the conversion math laid out anywhere.

2. **Rate limits** — I hit a 429 on my first burst of parallel calls. Is that per-key right now, and what’s the practical ceiling?

3. **Model choice for agent loops** — which combination works best for the planner vs. the cheap tool-call model? Looking for price/perf, not the flashiest.

P.S. Stev’s point on JAM + personhood at the end of the thread is a right call — from a practitioner side, model access and verifiable execution are what actually gate the agents, identity is what closes the loop. Would be great to see ComputeFlux sit at the model/execution side of that.

25

Thanks for trying the ComputeFlux beta and for the thoughtful feedback! Really appreciate you taking the time to test it and share these questions.

A few thoughts on each point:

1. Beta credits and pricing

Good point. We agree that users should have a clear understanding of how credits translate into actual inference costs, especially before running larger workloads.

We’re working toward more transparent, model-specific pricing and usage reporting, so developers can better estimate costs before running their applications. This is particularly important as we explore dotUSD payments for AI credits.

2. Rate limits (429)

Thanks for flagging this. The beta currently has capacity and rate-limiting constraints, particularly for bursts of parallel requests.

We want to make these limits clearer and improve the experience for developers running concurrent workloads. If you can share roughly how many parallel requests triggered the 429, that would be very helpful for troubleshooting.

3. Model selection for agent loops

Great question. We also believe that the most expensive model isn’t necessarily the best choice for every step in an agent workflow.

A practical approach is to use a stronger reasoning model for planning and a faster, lower-cost model for routine tool calls and execution steps.

One of our goals with ComputeFlux is to make it easier for developers to experiment with different model combinations through a unified OpenAI-compatible API.

4. JAM, Personhood, and verifiable execution

We agree that model access, verifiable execution, and identity could become important building blocks for autonomous agents.

Our near-term focus is on accessible AI inference, developer APIs, and dotUSD-powered payments. Longer term, we see opportunities for ComputeFlux to contribute to the AI model and compute execution layer alongside Polkadot-native infrastructure.

We’ve also recently submitted our Polkadot AI Adoption & dotUSD Utility Pilot proposal, building on the discussions in this thread.

Thanks again for the constructive feedback. We’d love to hear more about your agent setup and the models you’re experimenting with!

26

Update: ComputeFlux OpenGov Proposal Submitted — Referendum #1956

Hi everyone!

After months of development and community discussions, we’ve officially submitted our ComputeFlux: Polkadot AI Adoption & dotUSD Utility Pilot proposal to OpenGov.

:link: Proposal: ComputeFlux: Polkadot AI Adoption & dotUSD Utility Pilot

One important aspect we’d like to emphasize is our approach:

Web2 Experience. Web3 Benefits. No Wallet Required.

We believe mainstream adoption shouldn’t require users to understand blockchain before they can use a product.

With ComputeFlux:

  • Web2 users can sign up with email and start using AI services without a crypto wallet.

  • Web3 users can already log in through our integrated SubWallet.

  • Developers can access AI models through OpenAI-compatible APIs.

  • dotUSD integration is our next step, enabling users to pay for real AI computing services with Polkadot’s native stablecoin.

ComputeFlux is designed to be chain-agnostic, but we see a meaningful opportunity to bring new users and practical payment activity into the Polkadot ecosystem.

What We’re Proposing

A 3-month pilot requesting 16,384 USDC, with 50% allocated to AI usage credits and dotUSD payment pilot activities.

Our goal is not simply to distribute free credits. We want to demonstrate how accessible AI services can attract users, support developers, and create measurable demand for dotUSD payments.

We’d Love Your Feedback

We’re especially interested in the community’s thoughts on:

  • How can we make dotUSD payments useful for everyday AI users?

  • What would encourage Web2 users to explore Polkadot through AI services?

  • Which AI features or developer integrations would bring the most value to the ecosystem?

We’ll continue sharing progress, responding to technical questions, and reporting measurable outcomes.

Thank you to everyone who has contributed feedback and helped shape ComputeFlux so far!

Start with AI. Discover Web3 when you’re ready.

— ComputeFlux Team