`opencode` with Custom (Local) Models

Docs

Everything starts from the opencode.jsonc, stored in ~/.config/opencode/opencode.json. Coding also takes up a lot of tokens, it’s nice if you can get at least 60K context length.

Using Local/Custom Models

Ollama

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "llama2": {
          "name": "Llama 2"
        }
      }
    }
  }
}

The name of the model can be custom, but the model name as the key in the models dictionary must be the correct name specified in ollama.

[!NOTE] A standard installation (curl -fsSL https://ollama.com/install.sh | sh) of ollama would have the OLLAMA_CONTEXT_LENGTH set to 4096 (tokens), since that installation automatically puts ollama as a systemd service, go to /etc/system/systemd/ollama.service, and do the following: ``` [Unit] … (Leave as is)

[Service] Environment=OLLAMA_CONTEXT_LENGTH=XXXXX # Change “XXXXX” to something longer ExecStart=/usr/local/bin ollama serve … (Leave as is)

[Install] … (Leave as is)

[!NOTE] To immediately see the the new context length in effect, run the command sudo systemctl daemon-reload

LMStudio

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "lmstudio": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "LM Studio (local)",
      "options": {
        "baseURL": "http://127.0.0.1:1234/v1"
      },
      "models": {
        "google/gemma-3n-e4b": {
          "name": "Gemma 3n-e4b (local)"
        }
      }
    }
  }
}

Similar to ollama, just some value changes. (like in models, name, baseURL)

`llama.cpp` (& other custom providers, like `vllm`)

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "PROVIDER NAME": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "DISPLAYED PROVIDER NAME",
      "options": {
        "baseURL": "http://127.0.0.1:PORT/v1"
      },
      "models": {
        "ACTUAL MODEL NAME (e.g. google/gemma-3n-e4b)": {
          "name": "DISPLAYED MODEL NAME (e.g. Gemma 3n e4b)"
        }
      }
    }
  }
}

It’s really easy if you have an OpenAI API compatible server.

Serving/Attaching to other clients

Server

$ opencode web --port PORT --hostname 0.0.0.0

There is an optional --mdns option, which gives it a domain name by default opencode.local; change that value via --mdns-hostname.

Client Attachment to Server

TUI: opencode attach HOSTNAME
Web: Visiting that URL

#ai #ai/realworld

opencode with Custom (Local) Models