Understand what happened.
Follow session history, inspect tool outcomes, and see where an agent is waiting for input or approval. Make the next decision with the execution in view.
Nylorun
Observable. Portable. Composable.
Agent infrastructure built around your code. Define agent behavior, connect your tools, and run sessions you can inspect and control.
1import { Agent, tool } from "@nylorun/agents"2import { z } from "zod"3 4const lookupOrder = tool({5 name: "lookup_order",6 input: z.object({ orderId: z.string() }),7 async run({ orderId }) {8 return { orderId, status: "shipped" }9 },10})11 12export const assistant = Agent({13 id: "assistant", name: "Order assistant",14 instructions: "Use lookup_order for orders.",15 tools: [lookupOrder],16})Build agents you can understand, own, and evolve.
An agent needs more than a model call. It needs context across turns, tools that do real work, and a way to pause for a person. You define the behavior and business logic. Nylorun coordinates execution and records session progress.
Keep your code yours.
Keep agent definitions separate from hosting. Run your tools in your application environment, with the dependencies and business logic you already own.
Evolve the behavior.
Bring instructions, tools, capabilities, and hooks together. Reuse the pieces that work and extend an agent as your application grows.
Your agent's logic.
Nylorun's runtime.
Define your agent. Start a session through the API. Nylorun runs the model-and-tool loop, while your tools execute beside your data.
ONE SESSION, END TO END
From a question
to work done.
Send input from your app. The Runtime calls the model with your agent’s instructions.
support / session 4412
Illustrative preview
Your application
Your code. Your environment.
Customer message
Where is order #4412?
Tools beside your data
Nylorun Runtime
The loop runs here.
STEP 01
Decide the next action
The model requests order details.
Input → Runtime
EVERY STEP IN VIEW
- Input received: recorded
- Tool requested: upcoming
- Result received: upcoming
- Response ready: upcoming
Your tools execute in your environment. Their results return to the Runtime.
OSS Runtime/Cloud Runtime · Coming soon
Start locally. Choose how you run.
Build with Agents, Harness, and Runtime. Start with OSS Runtime and its Studio experience. Cloud Runtime is coming soon.
Agents
Define agents, work with sessions, and connect your application’s tools.
@nylorun/agentsHarness
The shared execution foundation that brings agent behavior and tools together.
@nylorun/harnessRuntime
Run sessions in your own environment with OSS Runtime. Cloud Runtime is coming soon.
Work with Studio during development.
@nylorun/runtimeThe open-source APIs are evolving during beta. Follow the documentation for current setup steps and package availability.
Same Harness. Same protocol. Managed at scale.
Cloud Runtime will run the same Harness and use the same protocol as OSS Runtime, so you can switch between self-hosted and managed execution without rewriting your agents. Cloud adds distributed infrastructure built for production scale and reliability, operated by Nylorun.
Apply to explore it with us. Applications are reviewed manually; access and approval timelines are not guaranteed.
Nylorun, in plain terms.
Your code, the runtime, and how to get started.
What is Nylorun?
Nylorun is agent infrastructure built around your code. You define agent behavior and business logic; the runtime coordinates model calls, tool actions, and session progress. The stack has three parts: Agents, Harness, and Runtime. Studio is part of the OSS Runtime experience; Cloud Runtime is coming soon.
Explore the documentationWhere does Harness fit?
Harness is the shared execution foundation: agent definitions, execution contracts, and the agent loop. Applications use Agents to author agents and work with sessions. OSS Runtime and the upcoming Cloud Runtime consume Harness independently.
Explore the open-source foundationWhere do my tools run?
Your tools and hooks run in a connected executor in your application environment. It receives actions from the runtime, runs your implementation, and returns results. Your application remains responsible for business permissions and access to its own services.
Connect your toolsWhat does Runtime manage?
OSS Runtime coordinates model calls, persists session state and history, and tracks actions and pending approvals or responses. You can inspect session progress and uncertain work. Recorded progress does not guarantee that every interrupted external operation can be safely repeated.
Read about OSS RuntimeHow do I develop locally?
Use the current setup guide to create a compatible project. Define your agents with Agents, then run OSS Runtime locally. Studio is part of that local experience: try conversations and inspect sessions as you develop. The documentation covers current package availability and setup steps.
Start buildingWhat is available today?
Agents, Harness, and OSS Runtime are experimental beta, with APIs and packaging still evolving. Follow the repository and setup documentation for the current version. Cloud Runtime is coming soon. You can apply for early access; applications are reviewed manually, with no guaranteed access or approval timeline.
Cloud Runtime statusYour next agent starts here.
Start with the open-source workflow. Bring your code, connect your tools, and see how your agent runs.