---
type: "post"
id: "copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b"
slug: "copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b"
title: "CopilotKit released OpenDuet , an MIT-licensed template for proactive, self-hosted AI agents."
description: "CopilotKit has published OpenDuet, an MIT-licensed open-source template for a proactive, self-hosted AI agent."
retrieval_nugget: "CopilotKit has published OpenDuet, an MIT-licensed open-source template for a proactive, self-hosted AI agent. The repository matters because it presents agent work as a visible product workflow—tasks, questions, channels, and intervention points—rather than as a single chat exchange."
published_at: "2026-10-28"
updated_at: "2026-10-28"
record_date: "2026-10-28"
date_kind: "published_at"
topics: ["anthropic"]
entities: ["CopilotKit","OpenDuet","OpenAI","Anthropic"]
related_patterns: ["harness-architecture-outlives-model-choice","ai-native-orgs-move-to-review-and-orchestration","agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "CopilotKit has released OpenDuet, a new MIT-licensed template for proactive, self-hosted AI agents; the archive covers its earlier OpenDots template."
prior_refs: ["memory:1b9de7e9-a068-8b54-ad28-647563ccb7b4","memory:94a20b05-5413-8611-a388-72816d78b848"]
source_url: "https://github.com/CopilotKit/OpenDuet"
source_urls: ["https://github.com/CopilotKit/OpenDuet"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b/","markdown":"https://newruntime.com/posts/copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b.md","json":"https://newruntime.com/posts/copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b.json"}
---

# CopilotKit released OpenDuet , an MIT-licensed template for proactive, self-hosted AI agents.

## Retrieval answer

CopilotKit has published OpenDuet, an MIT-licensed open-source template for a proactive, self-hosted AI agent. The repository matters because it presents agent work as a visible product workflow—tasks, questions, channels, and intervention points—rather than as a single chat exchange.

## Summary

CopilotKit has published OpenDuet, an MIT-licensed open-source template for a proactive, self-hosted AI agent. The repository matters because it presents agent work as a visible product workflow—tasks, questions, channels, and intervention points—rather than as a single chat exchange.

## What happened

CopilotKit released OpenDuet as an open-source template for “a proactive AI agent that gets your work done and keeps you in the loop.” The public repository is licensed under MIT. Its framing is important: OpenDuet is a template, not a hosted service or a finished autonomous-worker product. A team is expected to take the code, configure it, run it in its own environment, and adapt the workflow to its own use case.

The project’s public materials describe an interface built around more than a text prompt. Recent repository changes refer to onboarding, main screens, task creation, channels, task views, and question sheets. That is a concrete product choice: the agent’s activity is meant to have visible state and surfaces where a person can inspect or respond. The stated goal is proactive work while keeping the user involved, rather than hiding long-running work behind a chat transcript.

OpenDuet includes a quick-start path and model-provider configuration. The repository says an installation can use OpenAI through `OPENAI_API_KEY` and `model.yaml`. It also documents an Anthropic option using `BOT_PROVIDER=anthropic`, `ANTHROPIC_API_KEY`, and `BOT_MODEL`. The setup materials mention a fake Google flow with five steps. These details show that the template is intended to be configurable rather than fixed to a single model vendor, although the source does not claim that every provider or model has been tested equally.

The repository also exposes operational concerns that are easy to overlook in an agent demo. A recent change corrected the shutdown script so that it removes only bot computers belonging to the current installation namespace, instead of potentially removing containers from another installation on the same host. The same change refers to image build, boot, supervision, and secret-isolation checks. At the same time, the project explicitly says that no OpenDuet image or deployment guide has been published. The code is public and a quick start exists; production deployment remains work for the adopter.

The project is active rather than archival: the supplied repository view shows 75 branches, 69 commits, no tags, and commits dated October 25–27, 2026. Those are repository signals, not a guarantee of stability, support commitments, or production readiness.

## What the archive adds

This is a new event on a known open, self-hosted-agent thread. [Hermes Agent](https://newruntime.com/posts/nous-research-s-hermes-agent-mit-licensed-self-a50abd/) is another MIT-licensed, self-hosted agent project, while [Kwant](https://newruntime.com/posts/kwant-launches-as-an-open-source-self-hosted-8521c0/) approaches the space as a multi-agent runtime with handoffs, routines, and approvals. OpenDuet is distinct in its emphasis on a user-facing proactive-work interface.

## Trend signal

OpenDuet supports the pattern that [harness architecture outlives model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/). Its source exposes an application layer, workflow state, supervision, isolation concerns, and switchable OpenAI or Anthropic configuration. The meaningful artifact is therefore not just a selected model, but the runtime and interface that organize work around it.

## Why now

The public repository shows changes immediately before and after its public opening: documentation and setup were corrected, internal material was removed, and the project clarified both provider configuration and deployment limits. This appears to be an early public release moment, not a mature packaged distribution.

## Connections to other events

An earlier CopilotKit example paired LangChain Deep Agents with a live frontend and treated the agent as a process with structured state rather than a chat reply. OpenDuet extends that general product direction into a self-hosted template whose visible task and question surfaces are part of the core experience.

## Why it matters

For teams building AI products, OpenDuet offers a useful implementation reference for an uncomfortable but necessary idea: a proactive agent needs an operational interface, not merely a prompt box. When work can create tasks, ask questions, move through channels, and remain inspectable, a team has places to define ownership and review.

The repository is also a reminder that self-hosting shifts responsibility rather than removing it. Teams must choose a model provider, manage API credentials, understand containers and namespaces, and decide how their own users approve consequential actions. The documented concern about one installation interfering with another is precisely the kind of operational boundary that a production team should test early.

## Limits and uncertainty

The source is a GitHub repository and its recent commit history, not an independent evaluation. It does not establish task-success rates, security guarantees, supported deployment targets, cost, latency, or which real business systems the template can safely operate. It provides no ready-made OpenDuet container image or deployment guide. The repository’s activity, branch count, and CI references do not prove production readiness. Its MIT license governs the code, not any external model APIs, credentials, services, or deployment infrastructure an adopter chooses.

## Skill to build now

Take one agent workflow your team currently runs through chat and sketch its observable states before adding more autonomy. Define a task record, a question or approval state, a completion state, and the human who owns each transition. Then test whether a second local installation can start and stop without touching the first. That exercise makes workflow visibility and runtime isolation concrete requirements rather than product aspirations.

## Reader takeaway

OpenDuet is not evidence that proactive agents are solved. It is a practical open-source example of the surrounding product shape: configurable model providers, visible work state, human intervention surfaces, and operational isolation. Builders can use it as a starting point, but should budget for deployment and verification work.

## Open question

Can OpenDuet turn its repository template into a repeatable production deployment without losing the isolation and user-control properties it is trying to expose?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)