Installing Open Code

Most coding agent tools lock you into one provider, one model, one agent. Want to switch from Claude to Gemini or from GPT to a local model? You usually have to change tools entirely. That’s the exact problem Open Code was built to solve. In this post I’ll walk you through everything: from installation all the way to using skills, MCPs, agents.md files, and a personal workflow I use to manage my entire second brain with Open Code.

Head over to the Open Code installation page and grab the curl command. Copy that, paste it into your terminal, and hit enter. Once it finishes, you’ll probably need to restart your terminal session. The first time I ran open code I got a “command not found” error. So either restart the terminal or open a fresh one, then type open code and it should fire up.

To close it, just press Ctrl C. And if you want to use Open Code inside a specific project, you need to navigate there first. I always recommend opening the project in VS Code, then using the integrated terminal. That way you’re already inside the project directory. Then you just type open code and you can start working within that project right away.

Connecting to Any Model

One of the biggest draws of Open Code is that it isn’t tied to one provider. You can connect to practically any large language model. Type /connect and you’ll see a list of options.

There are basically two types of providers: subscription‑based and pay‑as‑you‑go. With subscriptions you pay a flat monthly fee and get unlimited usage. The pay‑as‑you‑go route uses API keys where you load credits and get charged for what you use.

Open Code offers its own options for both:

Open Code Zen, this is the pay‑as‑you‑go, API key approach.
Open Code Go, a subscription plan. For $5 a month you get access to models like K2, GLM‑4, DeepSeek V4 Pro, and other top‑tier Chinese models.

You can also bring your own keys from any provider. If you already have a ChatGPT Plus or Pro subscription, you can authenticate with that. Just choose the OpenAI option, copy the link it gives you, authenticate, and then you can use models like GPT‑5.1 directly inside Open Code. I did exactly that, selected GPT‑5.1, said “hi”, and got a response immediately.

Even if you don’t want to pay a cent, there are free models available. Scrolling through the list you’ll see options like DeepSeek V4 Flash Free. Just select it, hit enter, and you’re chatting with a free model inside Open Code.

Basic Usage and Project Scaffolding

When you launch Open Code, look at the bottom right, you’ll see a context window tracker showing how much context the current session is consuming. To start fresh, just type /new.

Let’s try scaffolding a project from scratch. I was inside an empty directory and told Open Code: “I want to scaffold a NestJS server application that prints hello world.” It went to work. If you ever need to interrupt the agent, double‑tap Escape. Then type “continue” to pick up where it left off.

Once the scaffolding finished, I opened a new terminal, ran npm run start:dev, and opened localhost:3000 in the browser. There it was: “hello world.” The entire source structure, controllers, services, main.ts, was all set up and ready to go.

Adjusting Effort and Running Multiple Sessions

By default, Open Code shows an effort level like “medium.” You can bump that up or down depending on how much processing you want to throw at a task. Type /variance and you’ll see options for low, medium, high, extreme high. I switched to high right away.

You can also run multiple sessions at once. I had one session focused on refactoring a messy project, while another session was working on something else entirely. This is handy when you don’t want one long conversation to slow down other tasks. Use /new to start a separate session and jump between them as needed.

Diving Into Skills

Skills are a big part of what makes Open Code powerful, similar to what you’d find in Claude Code or Cursor. If you want to install a skill, you can just ask the agent to do it.

In one example I told Open Code to install the “super power” skill. It went ahead and fetched it. I had to accept the permission prompt to let the agent run the installation, then it finished. Once a skill is installed, exit the current session and restart so Open Code picks it up. After restarting, type /kills and you’ll see the full list of available skills, including the newly installed one. You can then reference them directly in conversations.

Setting Up Your agents.md File

An agents.md file works like a system prompt for your entire project. It defines rules, context, and how the agent should behave. Without one, the agent doesn’t have a clear picture of your project. You can initialize it automatically by typing /init inside Open Code. The agent scans your repository and builds an agents.md file tailored to what you have.

Here’s an example of what ends up inside:

Repository rule, a high‑level description of the whole repo.
Agent rules, how the agent should respond: keep things concise in plan mode, ask clarifying questions, use sub‑agents for feature implementations, coordinate with other agents for reviews, and use a stronger model for complex tasks.
Folder map, a breakdown of each folder and what it contains.
Context rules, where context comes from (for example, synthesized from a raw folder) and how it all interacts.
Tasks and shortcuts, references to the skills available for different kinds of work.

The whole file stays under 50 lines but gives the agent a crystal‑clear understanding of the project. You can also add custom rules. For fun, I added “respond with emojis at least” to the agents.md, started a new session, said “hi”, and sure enough, the response came back with emojis. That’s how quick it is to tweak the agent’s personality or enforce behavior across all interactions.

My Personal Workflow: Managing a Second Brain with Open Code

This is where Open Code gets really interesting. I’ve been using it to manage my entire second brain. Think of it like querying a knowledge base that’s not just static notes but a set of AI‑powered advisors.

The setup is built around a “board members” skill. I scraped YouTube transcripts of people I look up to, pulled their thinking patterns, and turned those profiles into board member personas. When I have a decision to make or a problem to solve, I trigger the “ask the board” skill. Open Code loads those personas and gives me recommendations from each one.

One concrete example: I asked, “Based on our context about YouTube data, how can we achieve 20% growth for content creation?” The board members gave different opinions, each from their own angle. Another time I asked how to improve my school community. Again, multiple perspectives, each person highlighting different problems and offering specific advice.

What’s powerful about this isn’t just the answers; it’s that I have a second pair of eyes, or several pairs, for any decision. I don’t have to open different tools or switch context. I’m already inside Open Code, working with my project, and I can call on these board members as if they were part of the team. It keeps everything unified: my code, my knowledge base, and my advisory group.

Points clés à retenir (Key Takeaways)

Open Code breaks vendor lock‑in by letting you connect to any model, subscription, pay‑as‑you‑go, or free.
Installation is a single curl command; after a terminal restart you’re ready to go.
You can scaffold entire projects with natural language prompts, adjust effort levels with /variance, and run multiple sessions concurrently.
Skills can be installed and extended to give the agent new capabilities, similar to MCPs.
The /init command bootstraps an agents.md file that acts as a system prompt for your project, keeping the agent aligned with your rules and folder structure.
Beyond coding, Open Code can act as an interface to a personal knowledge base, enabling AI‑driven advisory workflows that bring multiple perspectives to your decisions.

The tool isn’t just a coding agent, it’s a hub where models, skills, and your own context all plug together. Once you spend a little time with it, you’ll wonder why you ever tolerated being locked into a single provider.