Today we are launching Alumia, the first commercial product built on top of our open-source agent infrastructure. Alumia is a platform for running autonomous AI agents that can interact with the real world — browse websites, fill forms, write and execute code, manage files, send emails, coordinate across systems. You describe what you need in plain language. The agent figures out how.
Alumia has been in private preview for about four months with a handful of design partners. The feedback from those sessions drove almost every decision about the shape of the product. We learned, for instance, that the first thing people want from an agent is not a new task — it is to automate something they already do by hand. We learned that the trust curve is slow: nobody hands an agent their email on day one, and that is correct. We learned that the hardest part of building a production agent is not the language model, it is the hundred small decisions about error handling, retries, and graceful failure.
The product is structured around three ideas. First, agents should do work, not talk about work. Alumia agents are optimized for completion, not conversation. You assign a task and the agent comes back with a result — a filled-out form, an organized inbox, a scheduled meeting, a summary of a research topic — not a transcript of the model's internal deliberations. Second, trust is earned incrementally. Every Alumia agent starts with read-only access and graduates to write access only after the user explicitly grants it. Third, everything is observable. You can see every action the agent took, every decision it made, and roll any of it back if something goes wrong.
Under the hood, Alumia is built almost entirely from the open-source packages we publish. The connector layer is the same @hasna/connectors library that anyone can install. The memory system is the same @hasna/mementos package. The task manager is @hasna/todos. The browser automation is @hasna/browser. Alumia is what you get when you stitch those pieces together into a hosted product with a clean interface and solid operational tooling on top. Everything we publish as open source runs in production for real users the moment it ships. That is by design — it keeps us honest about quality.
Who is Alumia for? The short version is: small teams that have more work than people. Ten-person startups, solo operators, small consultancies, independent shops. The larger the company, the more they tend to have existing internal systems an agent would need to plug into, which is a harder integration problem. The smaller the team, the more clearly the problem is "we don't have enough hours" — and that is exactly what agents solve.
Common early use cases we have seen: inbox triage, lead research, competitive monitoring, simple back-office automation (invoicing, expense categorization, bookkeeping prep), recurring reporting, and content distribution. None of these is novel individually — there are point solutions for each of them. What Alumia provides is a single platform where any new workflow is ten minutes of setup instead of a multi-week integration project.
Pricing is transparent and on the site. There is a generous free tier for anyone who wants to try it, paid plans scale based on agent-hours consumed rather than seats, and there is a self-hosted option for teams that need to keep everything on their own infrastructure. We do not have a sales team and do not plan to hire one. If you need help getting started, the docs are thorough and the support email reaches an engineer who uses the product daily.
A few things we want to be upfront about. Alumia is not perfect. The agents make mistakes. They occasionally get stuck. They occasionally misinterpret instructions. We have invested heavily in observability and rollback tools precisely because we expect errors and want you to catch them quickly. If you compare Alumia to a junior employee who is very fast but needs clear instructions, you will have roughly the right mental model. That is also, we think, the honest framing for where AI agents are right now industry-wide. Anyone telling you different is selling something.
For builders: Alumia is open to custom agent development. If you have a workflow that does not fit the standard patterns, you can write your own agent logic in TypeScript, use our SDK, and run it on the same infrastructure. There is a growing ecosystem of community-contributed agents — we link to all of them from the docs.
A thank you to the design partners who spent time with us during the preview. Several of them made product decisions that would have taken us much longer to figure out on our own. The resulting product is better because of them.
What is next. We have a full roadmap for the next six months that we will share on the blog as items ship. In the short term, the priorities are: better team collaboration features, an expanded connector catalog, and a marketplace where teams can share agent recipes with one another. If there is a feature you need, the feedback link in the app goes to a real person.
Try it at alumia.com. We think you will like it.