Exploring Knecht Cloud: Talking to an AI agent in GitHub issues – Part 2
In part one of this series, we installed Knecht Cloud on a Hetzner VPS. For this second part, I want to see how quickly I would be able to chat with an AI agent within my GitHub issues.
Spoiler: It took me a few minutes! 😮
Creating an (Open Code Go) API Key
First, I created an API Key for my OpenCode Go subscription in the OpenCode Go dashboard.
Open Code Go is the attempt to provide a low budget AI subscription option, currently for $10 a month. So it’s perfect for first experiments with Knecht Cloud.
I added the API key in the Knecht settings screen. As model, I selected „deepseek-v4-flash“ for now, since this is a very cheap / cost efficient model:
If you want to do more complex tasks, you could use glm-5.2 for example.
You can also use other AI API providers and models:
My first AI agent workflow with Knecht Cloud
I just created a very simple workflow called „GitHub Issue Conversation“ in the Knecht Cloud dashboard.
The important part is to select a trigger for newly created issues on GH:
You could also tie this a specific label of your choice only, e.g. „enhancement“.
By the way: Knecht has also support the events „(new) Pull Request“ and „Push“ for GitHub.
Next, I only added a simple „AI step“ to the workflow with a very basic prompt for now:
Follow the current conversation within the issue {{ inputs.url }}
Reply on the first run with „Hello, I’m ready! just mention me“ in the issue“
If you type „{{„, you can check out the variables you can add to your prompt:
Don’t forget activate the „Automation“ toggle …
… and that’s really it!
First chat with the agent
After creating a new issue on GitHub, a new run is created in the Knecht Cloud dashboard and the agent responds as desired:
By mentioning the displayed username of my Knecht instances bot, I can require further steps:
Beware: The bots name won’t show up automatically after typing „@“ in the issue comment, that is a GitHub limitation.
In the Knecht Cloud dashboard, you can analyze each triggered mention. The clever thing is, that this will all be executed within the same run, therefore the context is shared.
I can even ask the agent to create a new PR, if wanted.
And … I could do this all on my phone within the GitHub app as well. 🎉🥳 Huge kudos to Samuel (Reichör) for making this work!
You can find more complex examples in the official blog post:
https://knecht.works/updates/sessions-and-mentions
Analyzing the AI agent costs
What I like about OpenCode Go is that they have a very simple and straightforward interface. You can filter the usage cost by API key:
You can also use extra budget if the Go limits are reached:
Guess I don’t really need to mention it, but setting a monthly spending limit is recommended if you want to use Auto Reload 😉
„Follow up“ box in the Knecht dashboard
There is also the option to do a follow up prompt within the run event in the Knecht Cloud dashboard, if you don’t feel like commenting on GH:
Beware that the answer will be posted to the GH issue as well, because we activated „let the agent answer on the issue or PR“:
To be continued …
You can find more complex examples in the official blog post:
https://knecht.works/updates/sessions-and-mentions
Huge kudos to Samuel (Reichör) for making this work! 👏
Regarding this series, I’m not sure yet, what I want to try next.
If you have an idea, let’s chat at Knecht Discord!