Quorum

Landing
Agent Group "Personality" Selection
Agent Task Graph & Task Scheduler
Carbon Cost Savings Dashboard
Agent Group Output

Inspiration

AI keeps getting better, but we kept running into the same problem. After a run, most teams still do not really know if they made the best execution choice for cost and sustainability.

Watching everyday users select premium models for the most trivial tasks like writing emails or quick document summaries really pushed us to build this out. Collections of smaller, optimised models often perform just as well on the task for much less money and with less environmental impact.

We wanted model choices to feel practical and responsible, so quality, cost, and carbon are considered together in everyday decisions. This is the essence of Quorum.

What it does

Quorum runs complete AI workflows, not just single prompts. Your requests are broken into tasks and subtasks, which are then routed to the most suitable model. Our agent graph then visualises progress in subtasks in a clear visual timeline.

Alongside execution, Quorum handles the full economics of each run. Token usage is tracked in real time, micropayments are applied per run, user balances and Stripe top ups are managed in-app, and every transaction is written to a ledger for transparent reporting.

Sustainability is a key factor in the task delegation decision loop. Lower priority workloads can be scheduled for greener time windows, while carbon impact is tracked from inference data, and model routing balances quality, cost, and efficiency in real time. The result is one coherent platform that brings orchestration, observability, billing, and carbon aware optimisation into a single product experience.

How we built it

We built Quorum on a heavy Python backend that handles the core execution flow, task decomposition, agent orchestration, model routing, billing, Stripe top ups, balances, ledger tracking, and carbon estimation. A single request can be split into the right steps, run end-to-end, and measured for both cost and environmental impact.

On the frontend, we used Next.js and TypeScript to make everything visible and usable, including workflow graphs, agent nodes, execution timelines, scheduling controls, a workflow marketplace, and cash balance top ups. A clean API layer keeps backend decisions and frontend state in sync in real time. The hard part was making orchestration, payments, observability, and carbon aware scheduling work together in one smooth product experience.

Challenges we ran into

Our primary challenge was integrating orchestration, model routing, real time visualisation, micropayment billing, and carbon-aware scheduling into one smooth product. Maintaining a reliable ledger and Stripe flows through failures and retries, while still balancing quality, latency, cost, and prioritising sustainability in each decision was also a fun problem.

Product clarity proved to be a real thought experiment. The system is technically complex, so we thought hard on how to present Quorum plainly for users who really seek clear answers in understanding their workflows.

Accomplishments that we're proud of

We are seeing outcomes up to 7x cheaper at comparable quality by routing each task to the right models instead of defaulting to the most expensive option. Simple tasks go to smaller optimised models, while only complex tasks use premium models, which cuts waste without sacrificing output quality.

We also shipped a full end-to-end system that combines multi-agent orchestration, real time execution visibility, Stripe based micropayments, and ledger-level tracking. Users have clear, auditable run-level accountability while letting them optimise quality, cost, and sustainability in one workflow.

What we learned

We learned that smart routing makes a real difference on cost, but only if you keep evaluating performance continuously and stay transparent about why decisions are made. It is not enough to route once and assume it stays optimal.

We also saw that users trust the system much more when each run is clearly priced and tradeoffs are visible in real-time. The biggest lesson was around sustainability. It only works when it is part of the actual execution decision with quality and latency, not something shown later in a separate report.

What's next for Quorum

Next, we are improving adaptive routing and evaluation loops so decision quality gets better over time. We are also expanding enterprise capabilities such as team budgets, policy controls, and audit ready reporting.

On the product side, we plan to grow integrations, expand the workflow marketplace, and make Quorum easier to deploy in real production environments.