Launching today
Terse - Durable Actors
Infrastructure for Multiplayer AI Products
15 followers
Infrastructure for Multiplayer AI Products
15 followers
We provide an army of tiny servers to help you build incredible and cost efficient Multiplayer AI at scale. It's an open source alternative to Cloudflare's Durable Objects. Each resource or agent gets it's own micro-vm with a co-located SQLite DB. It can only process one request at a time making it concurrency safe. Onboard each one, is a websocket server. Allowing clients of any form to connect and listen for events on the resource.
Free Options
Launch Team / Built With
Render Sandboxes: a computer for every agentIsolated compute, persistent state and a real filesystem
Promoted
Hi Product Hunt!
We are building in the growing Actor space. As we move away from static, long lived UI to ephemeral real-time generated ones, the need for small pieces of compute that can spawn quickly and cheaply and be stored at little cost will skyrocket.
In addition, as more and more people use agents everyday and enterprises use more AI tools, collaboration will become paramount between humans and agents.
The problem? Our infra today wasn't built for this.
Central Websocket servers and ACID databases simply can't keep up with Agents that are creating content 90x faster than humans. As ultrafast models enter the market, that will increase it to 900x faster!!
The solution? Distribute the load. Have each agent write to it's own SQLite DB. Have each Agent serve it's own websocket server. Now your backend becomes stateless, simple lightning fast and cheaper to run. The fleet of Durable Actors handle real-time elements of your app.
Each actor comes into memory only when a request comes in. After 10s of inactivity, it hibernates into object storage costing you close to nothing.
Durable Actors are great at protecting resources shared by humans and agents. Think Figma canvas, an artifact (ex: Claude artifacts), a Word document. Clients can connect/subscribe to each of these, submit changes, read latest state and because each actor only processes one request at a time, you are protected against race conditions.
Our hosted version allows you and your team to use our declarative and type-safe API to define you actors. You can then test them locally with a development server. When it's ready, run terse deploy and your actors are now reachable via URL. Adding it to your existing project is trivial. Just run the generate command and it will build you a type-safe client with 0 dependencies.
Would love to know what you think! Feel free to reach out with any questions, bugs, concerns etc :)
founders@useterse.ai
hibernating into object storage after 10s idle is the bit that sold me, paying for a live websocket server per document nobody is looking at is what kills these setups
what happens when an agent hammers one actor with a burst of writes tho??? does it just queue them one by one or is there a point where you start rejecting, probaly matters a lot once agents are the main writers
@thamibenjelloun Yes! it queues! we make sure to log that like crazy so it's obvious when it happens.
The hibernate-and-wake flow makes reconnecting clients an interesting part of the experience. Showing which document version they resumed from would help make that transition easier to understand.