Phase 1 — OpenCode Server integration (Mode A)

[philbudden]

Phase 1 — OpenCode Server integration (Mode A)

Goal: Real LLM responses via OpenCode → Ollama

• Add opencode-server container to compose stack
• Implement server-mode.ts (OpenCode HTTP/WS API client)
• Wire gateway → runOpencodeAgent() → OpenCode Server → Ollama
• Stream response tokens back to Matrix thread
• Session state serialisation (transcript persistence)
• Handle OpenCode session errors, retries, timeouts

Exit criteria: Interactive Matrix conversation with Ollama-backed responses,
session continuity across multiple turns in same thread