Use Ollama to Power OpenCode

> Run models with Ollama locally and point OpenCode at Ollama's OpenAI-compatible API. Keep coding prompts on-device and avoid API spend.

Index

• Overview
• Setup
  • Start Ollama locally
  • Configure OpenCode
• Beginner usage
• Pro usage
  • Project-local config
  • Hybrid local-and-cloud setup
  • Operational tips
• Cost savings guide
• Privacy guide
• Security guide
• Appendix

Overview

OpenCode can connect to any OpenAI-compatible provider. Ollama exposes one at http://localhost:11434/v1, so you can run OpenCode against a fully local model while keeping the same TUI workflow, project rules, and file-editing flow.

Why Ollama for OpenCode?

• Simple local setup with one background server
• No per-token API cost after the model is downloaded
• Good fit for private repos or offline-ish workflows
• Easy model switching without rebuilding config from scratch

Setup

1) Start Ollama locally

ollama --version
ollama serve
ollama pull qwen3.5:3b

Quick API checks:

curl http://localhost:11434/api/version
curl http://localhost:11434/v1/models

If ollama serve is already managed by the desktop app or a service, you only need to make sure the server is listening on localhost:11434.

2) Configure OpenCode

Option A: Global config

Create or edit ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "qwen3.5:3b": {
          "name": "Qwen 3.5 3B (local)"
        }
      }
    }
  },
  "model": "ollama/qwen3.5:3b",
  "small_model": "ollama/qwen3.5:3b"
}

Option B: Project-local config

Create ./opencode.json in the repo you want OpenCode to use. OpenCode merges it on top of your global config, which is useful when only some projects should default to Ollama.

TUI flow

1. Launch OpenCode: opencode
2. Run /models
3. Select ollama/qwen3.5:3b

Beginner usage

1. Confirm Ollama is serving: curl http://localhost:11434/api/version
2. Confirm your model exists: ollama ls
3. Open your repo and launch OpenCode: opencode
4. Run /init if the project does not already have AGENTS.md
5. Use /models to select your Ollama-backed model
6. Start with a simple repo question or small edit request

Quick first run

# Terminal 1
ollama serve

# Terminal 2
cd your-project
opencode

Inside OpenCode:

• /models → choose ollama/qwen3.5:3b
• Ask: Summarize this codebase and point out the highest-risk area

Pro usage

Project-local config

Use a per-project opencode.json when you want one repo pinned to local models without changing your global default:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "qwen3.5:3b": { "name": "Qwen 3.5 3B (local)" },
        "qwen3.5:7b": { "name": "Qwen 3.5 7B (local)" }
      }
    }
  },
  "model": "ollama/qwen3.5:7b",
  "small_model": "ollama/qwen3.5:3b",
  "share": "disabled"
}

This gives you a larger default local model for coding and a smaller one for lightweight tasks.

Hybrid local-and-cloud setup

Keep a cloud model available for difficult refactors while making Ollama your cheap default:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "qwen3.5:3b": { "name": "Qwen 3.5 3B (local)" }
      }
    }
  },
  "model": "anthropic/claude-sonnet-4-5",
  "small_model": "ollama/qwen3.5:3b"
}

Use /models to switch when a task needs more reasoning depth.

Operational tips

• Keep Ollama and OpenCode on the same machine to minimize latency.
• Pull models before you need them so OpenCode does not stall on first use.
• Use smaller local models for repo exploration and repetitive edits.
• If responses suddenly fail, verify both http://localhost:11434/api/version and http://localhost:11434/v1/models.

Cost savings guide

• After the initial model download, local usage has no per-token API bill.
• Use Ollama for exploration, codebase summaries, and draft edits.
• Reserve cloud models for final review or harder architectural work.
• Set small_model to your Ollama model so cheap background tasks stay local.

Privacy guide

• Local Ollama keeps prompts and code on your machine by default.
• In OpenCode, set "share": "disabled" if you do not want session sharing available.
• Disable unused remote providers for stricter local-only workflows:

{
  "share": "disabled",
  "disabled_providers": ["openai", "anthropic", "gemini"]
}

• Remember that OpenCode project rules, transcripts, and repo files still live on your own disk.

Security guide

• Keep Ollama bound to 127.0.0.1; do not expose port 11434 to the public internet.
• If you ever proxy Ollama beyond localhost, add authentication and network controls in front of it.
• Treat local models as code-execution-adjacent infrastructure: review agent permissions and keep destructive actions on approval when working in sensitive repos.
• Re-check your active model with /models before making broad repo changes.

Appendix

Quick refs

ollama ls
ollama show qwen3.5:3b
curl http://localhost:11434/v1/models
opencode

See also: ./ollama-cli.md, ./ollama-desktop.md, ../opencode/opencode.md, and ../llamacpp/llamacpp-opencode.md.