Moving away from chat
So I wanted to move away from chat to a terminal-based interface to run more agents and eventually give control to my computer.
Options were:
| Tool | Style | Models | Open source? | My take |
|---|---|---|---|---|
| OpenCode | Terminal/TUI | GPT, Claude, Gemini, local models, etc. | ✅ | Best fit for what you’re describing |
| Aider | Terminal | Many providers/local models | ✅ | Excellent if you want a very Git-centric workflow |
| Codex CLI | Terminal | OpenAI models | ✅ | Very capable, but less model-agnostic |
| Claude Code | Terminal | Claude | ❌ | Excellent agent, but not open source |
| Cline | VS Code | Many models | ✅ | Better if you still want an IDE |
I wanted to try the open-source models, so I decided to go ahead with OpenCode.
Quick setup instructions
-
Ensure you have npm and Node.js installed.
To check:
node --version npm --version -
Next, install OpenCode:
npm install -g opencode-aiThis means:
npm→ Node’s package managerinstall→ install something-g→ install it globally, soopencodeis available from any directoryopencode-ai→ the OpenCode CLI package
To check:
opencode --version -
Once you have it installed, you can close PowerShell and open a new one.
-
Launch OpenCode:
opencode -
This should open the OpenCode interface.
-
To connect to the model:
/connect -
Again, we have multiple options now:
- You can try ChatGPT if you already have Plus or Pro
- Since I wanted to try open-source models, I decided to set up OpenRouter:
Provider Top Models Quota / Limit RPM Context Window Best For Google AI Studio Gemini 3.6 Flash 1,500 req/day 15 1M tokens Long-context analysis & massive codebases Groq Llama 3.3 70B, Mixtral 1,000 req/day 30 128K tokens Real-time apps & fast agentic loops Cerebras Llama 3.3 70B ~1M tokens/day 30 1M tokens Batch processing & high throughput GitHub Models GPT-4o, Claude 3.5 Sonnet, DeepSeek R1 50-1,000 req/day 15 8K-128K tokens Occasional frontier model access Mistral Codestral ~1B tokens/month Varies 32K-256K tokens Heavy coding workloads & high volume OpenRouter Space Bunny, Laguna, North Mini Code 50 req/day 20 Varies Model routing & experimentation -
I decided to create an OpenRouter account and get the API key to use it in the OpenCode setup.
-
Now, if you do not have any credits, you will be restricted to 50 requests per day, which I exhausted pretty soon, so I recommend adding $10 in credits so that you do not have this restriction.
You should be all set up!
I started with commands like cleaning up my disk space, setting up my git repo locally, etc.! Will add more things soon!
Originally published on Substack.