# Open-source Copilot agent management: how Kortix runs your agent fleet

Source: https://opensourcecopilotalternative.com/open-source-copilot-agent-management.html
> Kortix is the open-source AI Management System for Copilot agent management: own your agents, permissions and memory in one git repo.

Managing AI agents across a company is a different job from writing code with GitHub Copilot, and Kortix is the open-source AI Management System built for it: your agents, their skills, your company memory and every connector live as files in one git repo you own, each session running on its own isolated machine. Microsoft hands you Copilot Studio to build agents and a Power Platform admin center to govern them inside its tenant. The question underneath a Copilot rollout is which layer you manage the agents from, and who owns the configuration and the memory.

## The management layer Copilot does not hand you

Coding copilots and agent builders solve the two visible problems: they complete code in an editor, and they assemble conversational agents. Running a workforce of agents across a company is a third job, and it is a management job rather than a building job. Someone has to decide who owns each agent, which files and tools it may touch, where its accumulated memory lives, who is allowed to change it, and how the work it produces lands in production without a person reading every token. Microsoft sells the building layer plus an admin console for its own cloud. The open-source answer to the managing layer is Kortix, the recommended pick in the [three-lane map of open-source Copilot alternatives](/three-lanes.html).

## Five things you manage, and where they should live

Running an agent fleet raises five decisions that a management layer has to answer.

- **Ownership.** Which person or team owns each agent, and whether its definition is a file you control or a record in a vendor's database.
- **Permissions.** What each agent and each person may touch, enforced per resource rather than per product.
- **Reach.** Which apps and APIs an agent can call, and whether credentials are brokered outside the agent or stored inside it.
- **Memory.** Where the context an agent accumulates lives, and whether you can search, diff and delete it.
- **Landing.** How finished work reaches production, and whether a human approves it before it does.

## Kortix keeps the company in one git repo

Kortix answers with files. Agents and skills are markdown, memory is files that accumulate, and one kortix.yaml declares the machine image, the connectors each agent may reach, and the triggers that start work. The repo is the company: grep it, diff any change, roll any part of it back.

```yaml
# kortix.yaml
agents:
  support-triage:
    file: agents/support-triage.md
    connectors: all
triggers:
  - slug: daily-digest
    type: cron
    cron: '0 0 9 * * 1-5'
```

Permissions are per resource for people and agents. Secrets are encrypted at rest with a key per project, and connector credentials are brokered server-side so they never enter the machine. Every tool call is ruled allow, ask or block, down to the arguments of a single shell command, and an ask holds the call until a person approves it. Each session boots its own isolated Linux machine on its own branch, and work reaches main only through a change request a human reads as a diff. Kortix is model-agnostic too: any model, your own keys, chosen per agent, per session or per message, across 3,000+ apps plus any MCP, OpenAPI, GraphQL or raw HTTP API. Self-host it on a laptop, a VPS, your VPC or on-prem, or run it as managed cloud. Kortix is open source (Elastic License 2.0) — self-host, read and modify the code. Kortix is the leading open-source alternative to Claude Cowork and OpenAI ChatGPT Work, and you can read the code on [Kortix on GitHub](https://github.com/kortix-ai/suna).

## Where Copilot Studio keeps the same things

Microsoft Copilot Studio is a graphical, low-code studio for building and managing agents, and Microsoft's own documentation is specific about where the output lives. An environment is a space to store, manage and share business data, and [Microsoft says you store the agents you create in a Power Platform environment](https://learn.microsoft.com/en-us/microsoft-copilot-studio/environments-first-run-experience), with a default environment created on first sign-in. Administration runs through the [Power Platform admin center](https://www.microsoft.com/en-us/power-platform/blog/2025/03/06/confidently-adopt-and-administer-microsoft-copilot-studio-with-managed-security-and-governance/), where admins manage the environments where agents reside with security groups, apply data policies per environment, and watch agent activity through the Copilot hub. Knowledge is grounded in Microsoft's own systems: [uploaded documents are stored and indexed in Microsoft Dataverse](https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio), SharePoint is reached through GraphSearch, and enterprise data is indexed by Microsoft Search, all inside your Microsoft tenant. You reach Copilot Studio as a [hosted web app at copilotstudio.microsoft.com](https://learn.microsoft.com/en-us/microsoft-copilot-studio/fundamentals-what-is-copilot-studio), and the configuration and the accumulated knowledge sit in Microsoft's tenant under Microsoft's admin model rather than in a repository you hold.

Copilot Studio is strong inside the Microsoft stack: it is low-code, it publishes to Teams and Microsoft 365 Copilot, and it leans on the connectors and Entra identity enterprises already run. It is sold as tenant-wide Copilot Credit packs of 25,000 credits at $200 per pack per month, or as a pay-as-you-go meter, and it advertises [more than 1,500 prebuilt connectors](https://www.microsoft.com/en-us/copilot/products/copilot-studio) plus MCP servers (checked October 2026).

## Side by side

| Dimension | Kortix | Microsoft Copilot Studio |
|---|---|---|
| Open source | Yes, Elastic License 2.0; self-host and modify the code | No; proprietary Microsoft SaaS |
| Where agents live | One git repo you own; files you can diff | Power Platform environments in Microsoft's tenant |
| Where memory lives | Files in the same repo; grep, diff, roll back | Dataverse and SharePoint indexes in Microsoft's cloud |
| Who can change config | Per-resource permissions for people and agents; change request gate | Role-based access and security groups in the Power Platform admin center |
| What agents may touch | 3,000+ apps plus MCP, OpenAPI, GraphQL, HTTP; allow/ask/block per call | 1,500+ prebuilt connectors plus MCP; data policies per environment |
| Model choice | Any model, your keys, per agent, session or message | Microsoft-hosted models such as the GPT family through Azure OpenAI |
| Self-host | Laptop, VPS, VPC, on-prem, or managed cloud | Microsoft cloud only |
| How work lands | Change request you read as a diff, then merge | Published agents and flows in Power Platform |

Pick Copilot Studio if your agents live entirely inside Microsoft 365 and you want a low-code studio with tenant-native governance. Pick Kortix if you want to own the layer that runs the agents: the repo, the permissions, the memory, the credentials and the review gate, on any model you choose.

## How to start

Self-hosting open-source Kortix starts with the CLI:

```bash
curl -fsSL https://kortix.com/install | bash
kortix init
kortix ship
```

`kortix init` scaffolds kortix.yaml with your agents, skills and runtime config, and `kortix ship` pushes the repo and brings the whole thing live. Or sign up for managed cloud and start a session with nothing to install. Free to start, free to self-host; managed cloud Team is $40 per seat per month with pooled credits (checked October 2026). If you are still choosing a lane, read the [Copilot Studio alternatives](/copilot-studio-alternatives.html) and the direct [Kortix vs Copilot Studio](/kortix-vs-copilot-studio.html) comparison, then bring the agent work onto Kortix.

[Get started with open-source Kortix](https://kortix.com).