English | 简体中文
loopd is a collaboration platform where people, Operators, and Agent execution
services (Harnesses) exchange persistent messages. Build Kubernetes Operators with
the Go loop-runtime toolkit to connect these participants with business systems,
from routing a question to running a long-term task.
An Agent can execute work. A business still needs to define its goals, coordinate participants, and decide when the result is good enough. loopd gives developers a collaboration platform and an Operator toolkit for expressing those decisions in ordinary code:
Loop = Resource(spec + status) + Reconcile
Resources hold goals and observed state; Reconcile decides the next step.
Each Operator defines its own process, collaboration model, and completion criteria. Server manages shared messages and includes a Harness Engine to drive calls through adapters for different Harnesses. The embedded loop-runtime toolkit lets Operators use these capabilities through verbs such as Speak and Prompt.
- Develop your own orchestration. Combine Agent judgment with deterministic code and business APIs. The Go loop-runtime toolkit provides shared capabilities for calling Harnesses, asking people, and publishing progress.
- Collaborate while work continues. People, Operators, and Harnesses exchange persistent messages. Users can add context during execution, and Operators can share partial results or request confirmation as needed.
- Keep long-running work observable. Conversations show shared progress and results; domain resources retain business state. Operators can coordinate planning, execution, and verification over time.
The bundled Router plans a request, runs independent Harness calls in parallel, and synthesizes an answer. Here it introduces Liu Bei, Guan Yu, and Zhang Fei; the detail panel shows the planning, execution, and synthesis alongside the answer.
For longer tasks, the LongHorizon Operator combines a Manager, Executor, and Auditor to plan CLI work, execute it, check artifacts, and decide what to do next.
Follow the Kubernetes Quick Start to install the server, Router, and Web UI. Configure an OpenAI-compatible model endpoint and credentials, open the UI, select Router, and submit a question.
The Quick Start uses temporary storage and an in-process AgentGo demo. Recovery across restarts requires persistent storage, saved Operator progress, and a Harness with persistent execution and an adapter that supports recovery; see the recovery contract.
To build an Operator, start with the Router source and runtime guide. For Harness integration and execution, see the Harness Engine guide. Explore the kernel, component stack, or image build guide for more detail.
