Table of contents
Step 1: Install OpenCode
Our agents work as an OpenCode plugin. You can use your own API keys for LLMs, i.e. it will work with a Chat GPT pro subscription.
Install instructions for OpenCode are here: https://opencode.ai/download
Step 2: Generate an API Key on our website
Create an API key at: https://platform.ncompass.tech
Step 3: Update ~/.config/opencode/opencode.json with the following contents
{
"$schema": "<https://opencode.ai/config.json>",
"plugin": [
[
"@ncompass/opencode-plugin-gpu-dev",
{
"ncompassApiKey": <API Key You Generated>
}
]
],
"mcp": {
"knowledge_bank": {
"type": "remote",
"url": "<https://knowledge-base.ncompass.tech/mcp>",
"headers": {
"x-api-key": <API Key You Generated>
},
"timeout": 30000
}
}
}
Step 4: Start OpenCode
You should see our knowledge_bank MCP in OpenCode and if you move through Tab modes, you should see Nc-Ask, Nc-Plan and Nc-Build popup.
Overall, our agent will automatically run appropriate profilers, analyze the resulting artifacts and suggest bottlenecks and optimizations that are missed by Claude Code / Codex / Cursor by default. We currently support nsys , ncu , torch.profile and pyspy .