The model call is the easy part. What breaks an agent in a real product are ordinary moments: the customer closes the tab mid-answer, a deploy goes out mid-Turn, a tool errors, the agent asks a question and the person answers tomorrow. Each one looks like an edge case. Together they are the difference between a demo and a product.
Nvoken exists so the founders and small teams actually shipping those agents get that part right without building it themselves.
Why Nvoken
We spent eighteen months building agent products at Deep Noodle and hit every one of those moments. The frontier labs handle most of them inside their own apps, with very large teams. Provider SDKs and agent frameworks cover parts. The all-in platforms cover more, and keep you inside their ecosystem.
Teams usually start with a provider SDK and a short model loop. Then the product asks for conversation history, customer-specific context, authorized actions, streaming, recovery, model choice, usage attribution, and spending controls. Each request is reasonable. The collection is an agent platform the team did not set out to build.
Nvoken gives those capabilities one contract and one operating model, hosted, and available for self-hosting if you contact us. The team spends its time on the behavior and experience that make its agent worth using.
Questions worth asking first
Most of these decisions get made by accident, right before launch, when the architecture can no longer change. We think a product team should answer them before the first line of agent code.
- Does the agent’s state need to persist between visits?
- If the customer refreshes, closes the tab, and comes back, should they rejoin where they left off?
- Should the agent keep working after the customer closes the app?
- When the agent needs to ask the customer something, what happens if the answer takes a while?
- When a tool call fails, does the agent keep going and get to decide whether to retry?
- When engineering deploys a new version, do in-progress agents lose their state?
Nvoken’s design starts from these questions: durable Turns, Conversations and MemorySpaces your application selects, waiting that holds no worker, and reconnect by ID from wherever the work was.
What it is
Nvoken is an AI agent platform that combines a full agent SDK with a hosted execution system. An agent, in this context, is software a person can converse with that can use context, decide what to do next, call tools to read or change product data, and work through multiple steps to a result.
Nvoken runs durable Turns, keeps the Conversations and MemorySpaces your application selects, coordinates the model-and-tool loop, waits for a person or a tool without holding a worker, streams progress a client can resume, records actions and usage, and routes across supported model providers.
How it fits
Your application still owns the customer experience, users, product data, authorization, billing, business rules, and the tools an agent is allowed to use. Nvoken runs the model-and-tool work and preserves only the continuity and memory your application explicitly selects.
That boundary is deliberate. Nvoken should make a product team more capable without asking it to hand over the parts of the product that make the business distinct.
An open-source foundation
Who makes it
Nvoken is a product of Deep Noodle, a bootstrapped technology company founded by Curtis Myzie. Curtis has spent more than 20 years building production systems and has served as a CTO, VP of Engineering, and VP of Product at venture-backed startups.
Deep Noodle builds and operates its own SaaS and agent infrastructure products, contributes open-source software, and helps other teams ship secure, durable systems. Questions about Nvoken are welcome at support@nvoken.com.