---
type: "post"
id: "mainplane-is-a-new-open-source-mit-licensed-a73b75"
slug: "mainplane-is-a-new-open-source-mit-licensed-a73b75"
title: "Mainplane is a new open-source, MIT-licensed agent harness written in Go."
description: "Mainplane is an MIT-licensed, open-source agent harness with a self-hosting quickstart for running agents across computers, connectors, model providers, files, browsers, and applications."
retrieval_nugget: "Mainplane is an MIT-licensed, open-source agent harness with a self-hosting quickstart for running agents across computers, connectors, model providers, files, browsers, and applications. For teams building agentic workflows, it is another attempt to make the harness—not a single model chat—the operational layer where access, context, and execution are organized."
published_at: "2026-10-31"
updated_at: "2026-10-31"
record_date: "2026-10-31"
date_kind: "published_at"
topics: ["agent-harness","claude","claude-code"]
entities: ["Mainplane","mainplane-ai/mainplane","Anthropic","OpenAI","Google","Claude Code","ChatGPT Codex"]
related_patterns: ["harness-architecture-outlives-model-choice","agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Mainplane adds a distinct MIT-licensed agent harness written in Go to the archive’s coverage of open-source agent tools."
prior_refs: ["memory:1d940e02-c1d7-81df-a69f-b86244fbd19b","memory:d003d0fc-b783-88b8-a8d9-bd96c30b57f4"]
source_url: "https://github.com/mainplane-ai/mainplane"
source_urls: ["https://github.com/mainplane-ai/mainplane"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:mainplane-is-a-new-open-source-mit-licensed-a73b75"
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/mainplane-is-a-new-open-source-mit-licensed-a73b75/","markdown":"https://newruntime.com/posts/mainplane-is-a-new-open-source-mit-licensed-a73b75.md","json":"https://newruntime.com/posts/mainplane-is-a-new-open-source-mit-licensed-a73b75.json"}
---

# Mainplane is a new open-source, MIT-licensed agent harness written in Go.

## Retrieval answer

Mainplane is an MIT-licensed, open-source agent harness with a self-hosting quickstart for running agents across computers, connectors, model providers, files, browsers, and applications. For teams building agentic workflows, it is another attempt to make the harness—not a single model chat—the operational layer where access, context, and execution are organized.

## Summary

Mainplane is an MIT-licensed, open-source agent harness with a self-hosting quickstart for running agents across computers, connectors, model providers, files, browsers, and applications. For teams building agentic workflows, it is another attempt to make the harness—not a single model chat—the operational layer where access, context, and execution are organized.

## What happened

The public Mainplane repository describes the project as “an open-source agent harness” and labels it “The open-source agent ecosystem.” It is publicly available under the MIT license. The repository contains Go module files and a server-oriented installation path, while its README positions the product as infrastructure for operating agents rather than as a standalone model or a prompt library.

The self-hosting quickstart instructs an operator to install and run `server quickstart` on an always-on server machine. The README says that setup walks the operator through adding a model provider, workers, and connectors. In Mainplane’s terminology, workers are computers that agents control; connectors are ways to message agents. The server is stated to run on macOS, Windows, and Linux. Hosting shared files has an additional constraint: it requires a compatible connected Linux device.

Mainplane says it is model agnostic and can use models from providers including Anthropic, OpenAI, and Google. Its FAQ also says a user can use the regular limits of a Claude Code subscription or a ChatGPT Codex subscription, or use an API key with any model. That is an access claim worth separating from a claim of model equivalence: the supplied material says these subscription limits can be used, but does not specify which capabilities, rate limits, or account terms apply in each configuration.

The project’s main product claims concern the runtime around those models. Its agents are described as having a restricted set of tools and managing their own context by retrieving key information from other sessions. Mainplane also says it links agents and computers with fine-grained security controls, without requiring Tailscale. It offers a shared remote filesystem intended for concurrent use without de-synchronization, plus native computer use for real browsers and applications.

The README says no data is shared by default; users may opt in to sharing data to improve Mainplane through settings. It also says a single device is sufficient. Community participation is unevenly open at present: feedback is welcomed as GitHub issues, but pull requests are open only to collaborators. The repository page listed 561 stars, 38 forks, one watcher, and 223 commits in the supplied snapshot. Those numbers describe a moment in a repository’s visible activity, not adoption, reliability, or production readiness.

## What the archive adds

This is a new event on a known harness story. [Mirror CLI](https://newruntime.com/posts/open-sourced-mirror-cli-the-rust-core-agent-6deb32/) framed the coding-agent harness as an operational surface developers must understand. [Hermes Agent](https://newruntime.com/posts/nous-research-s-hermes-agent-mit-licensed-self-a50abd/) and [SuperPlane v1.0.0](https://newruntime.com/posts/superplane-v1-0-0-first-stable-release-of-75d162/) likewise documented open-source agent systems, but they are separate projects. Mainplane adds a distinct self-hosted harness focused on controlling computers, files, connectors, and model access.

## Trend signal

Mainplane supports the pattern that [harness architecture outlives model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/). Its README treats the model provider as interchangeable while making workers, connectors, context retrieval, filesystem coordination, and computer-use controls first-class components. It also partially supports [agent security moves to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/), although the repository supplies claims about fine-grained controls rather than a technical security model or independent evaluation.

## Why it matters

A team moving from an assistant in a browser to agents that act across browsers, applications, and machines needs more than a model selection layer. It needs explicit answers to where execution happens, which device an agent can reach, how files are shared, how messages arrive, and what information is carried between sessions. Mainplane packages those concerns into one deployable harness and offers a route for teams that prefer self-hosting.

The attractive part of the design is also the part that needs the most operational scrutiny. A shared filesystem and real computer control can remove friction from multi-agent work, but they increase the importance of permissions, worker isolation, secret handling, recovery behavior, and auditability. Model-agnostic wiring can reduce provider lock-in; it does not by itself make the surrounding workflow portable or safe.

## Limits and uncertainty

The supplied source is Mainplane’s own repository and README. It does not provide a release date, benchmark results, security architecture, pricing, service-level commitments, supported connector list, or comparative evidence against other harnesses. “Fine-grained security controls,” safe concurrent file use, and native computer use are product claims, not independently verified outcomes. The source also does not establish maturity, production deployments, contributor governance, or the precise conditions under which Claude Code and ChatGPT Codex subscription limits work. Teams should treat the project as software to evaluate in a constrained environment, not as evidence that autonomous computer control is ready for unrestricted deployment.

## Skill to build now

Create a small harness acceptance test before adopting any self-hosted agent runtime. Use one disposable worker, one non-sensitive folder, and a bounded browser task. Record the permissions granted, the files created or changed, the context retrieved from prior sessions, and the recovery path after a failed action. This makes the relevant comparison concrete: not which harness has the longest feature list, but whether its execution and authorization boundaries are visible enough for your team to operate.

## Reader takeaway

Mainplane is an early open-source option for teams that want to self-host the layer connecting agents to models, workers, files, messages, browsers, and applications. Its model flexibility is useful, but the evaluation should begin with runtime boundaries and operational controls—not with the claim that any model can be plugged in.

## Open question

Can Mainplane demonstrate permission, audit, and recovery mechanisms detailed enough for teams to safely use its shared files and computer-control features beyond an isolated pilot?

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: [Agent Harness](https://newruntime.com/topics/agent-harness.md): Explore the agent-harness topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code 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)