Local models

Local models run through a bundled, pinned OpenCode harness — no account, no quota, nothing leaves your Mac. Point it at whatever serves an OpenAI-compatible endpoint: Ollama, LM Studio, or a llama.cpp server.

1. Run a model server

   Ollama is the quickest path:

   $brew install ollama && ollama serve

   Then pull a coding-capable model

   $ollama pull qwen3:14b

2. Enable Local in Synergoi

   Open Settings › Providers › Local models, choose the runtime, and pick the model from the list Synergoi reads from the server.

3. Pick it in the composer

   The provider pill lists your local models next to the cloud ones. Everything else — worktrees, approvals, Changes — works the same.

Context size matters

Agentic work needs room. Synergoi reads the model's real context window from the server and marks anything under 16k tokens as unfit for agent turns; you can still send, with a warning. 32k or more is comfortable for most repos.

What Synergoi turns off in OpenCode

To keep a local model on task the harness runs with OpenCode's automatic titling, summaries and compaction disabled, and with permissions set to ask so the app's broker stays in charge. If you want a different OpenCode build, set SYNERGOI_OPENCODE_BIN to its path before launching the app.