gratefulagents

Positioning

The self-hosted alternative to Devin and GitHub Copilot

GratefulAgents is an open-source coding agent platform that runs inside your own Kubernetes cluster. Devin and GitHub Copilot's coding agent are cloud services: the checkout, the agent process, and the credentials live on the vendor's infrastructure. GratefulAgents puts all three on yours.

Run it locally → Why self-host?

What changes when you self-host

The difference is architectural, not a feature checklist. On a cloud harness your source code, your prompts, and your model credentials all transit the vendor's network and execute on the vendor's machines. Self-hosting moves that boundary: agents run in sandbox pods in your cluster, against secrets you hold, and every run leaves a trace you can inspect.

Dimension Devin Copilot coding agent GratefulAgents
Where the agent runs Cognition's cloud GitHub's cloud Your own Kubernetes cluster
Where the repository is checked out Cognition's infrastructure GitHub's infrastructure A sandbox pod inside your cluster
Who holds the model credentials Cognition GitHub / Microsoft You — stored in your own workspace
Model choice Cognition's own models GitHub's supported catalogue Claude, OpenAI, OpenRouter, Grok, Copilot — your keys
Licence Proprietary SaaS Proprietary SaaS AGPL-3.0 open source
Who operates it The vendor The vendor You — Helm chart on Kind or k3s
Per-run observability The vendor's session view Agent session logs Traces, cost, tokens, tool calls, subagent graphs
Trigger surfaces Web UI, Slack GitHub issues and pull requests GitHub, Linear, Slack, cron
Client applications Web GitHub web Web, desktop (macOS, Linux), iOS, Android

GratefulAgents is in early development. It works for evaluation and internal team use, but expect bugs and breaking changes between releases, and note that high availability, managed hosting, and a support SLA are not documented.

Frequently asked questions

Is GratefulAgents a self-hosted alternative to Devin?
Yes. Both run autonomous coding tasks against your repositories. The difference is where that happens: Devin executes on Cognition’s infrastructure, while GratefulAgents executes in sandbox pods inside a Kubernetes cluster you run, using model credentials you hold.
Is GratefulAgents a self-hosted alternative to GitHub Copilot's coding agent?
Yes. Copilot’s coding agent is a cloud feature of GitHub.com, so the checkout and the agent process live on GitHub’s infrastructure. GratefulAgents runs the same class of work inside your own cluster, and can be triggered from Linear, Slack, or a cron schedule as well as from GitHub.
Does my source code leave my network?
The checkout, the sandbox, the tool calls, and the run history stay inside your cluster. Inference is the exception: prompts and the repository context included in them are sent to whichever model provider you configure, under that provider’s terms. If you cannot send code to a third party at all, you need a self-hosted inference endpoint as well.
Which models can I use?
Any provider whose credentials you store in your workspace: Claude (Anthropic), OpenAI, OpenRouter, Grok (xAI), and GitHub Copilot. Model choice is configurable per project and per role, so you are not tied to one vendor’s catalogue.
What do I need to run it?
A Kubernetes cluster. The Kind guide stands one up on a macOS or Linux laptop for evaluation; the k3s guide covers a persistent install on a fresh Debian or Ubuntu server. Both are installed with the same Helm chart.

For the architecture and requirements behind self-hosting, see the case for running coding agents in your own cluster. To try it, the Kind quick-start takes about fifteen minutes on a laptop.

Devin® is a trademark of Cognition AI, Inc. GitHub® and GitHub Copilot® are trademarks of GitHub, Inc. All trademarks are the property of their respective owners. This page describes the authors' understanding of each product at time of writing; capabilities change frequently. Consult each vendor's own documentation for authoritative information.