light-heart-labs--ods
wehub-resource-sync
91 行
3.3 KiB
Markdown
91 行
3.3 KiB
Markdown
# OpenCode
|
|
|
|
AI-powered coding assistant accessible via browser in ODS
|
|
|
|
## Overview
|
|
|
|
OpenCode provides a browser-based AI coding environment that integrates with your local LLM. It gives you code completion, explanation, refactoring, and generation capabilities through a web interface — no IDE plugin required. Unlike other ODS services, OpenCode runs directly on the host system (not in Docker) as a systemd-managed process.
|
|
|
|
## Features
|
|
|
|
- **Browser-based IDE**: Access your AI coding assistant from any browser on your local network
|
|
- **Local LLM backend**: All code and queries stay on your machine — connects to llama-server for inference when available
|
|
- **Code completion**: Context-aware completions for multiple languages
|
|
- **Explanation and refactoring**: Ask the LLM to explain, refactor, or debug your code
|
|
- **No IDE installation required**: Works entirely in the browser
|
|
|
|
## Deployment Type
|
|
|
|
OpenCode runs as a **host-level systemd service** (`type: host-systemd`), not as a Docker container. This means:
|
|
|
|
- It is managed by systemd on the host, not by Docker Compose
|
|
- There is no `container_name` — use systemd commands to manage the process
|
|
- LLM features require either OpenCode's own backend or llama-server to be running (either satisfies the requirement)
|
|
|
|
## Configuration
|
|
|
|
| Variable | Default | Description |
|
|
|----------|---------|-------------|
|
|
| `OPENCODE_PORT` | 3003 | Port for the OpenCode web interface |
|
|
|
|
> **Note:** LLM-powered features (completion, explanation) need either OpenCode's own backend or `llama-server` running. OpenCode itself starts independently.
|
|
|
|
## Accessing OpenCode
|
|
|
|
Once running, open your browser and navigate to:
|
|
|
|
```
|
|
http://localhost:3003
|
|
```
|
|
|
|
## Requirements
|
|
|
|
| Requirement | Details |
|
|
|-------------|---------|
|
|
| GPU VRAM | 8 GB minimum (for LLM inference via llama-server) |
|
|
| LLM service | `llama-server` must be running |
|
|
| GPU backends | NVIDIA, AMD |
|
|
|
|
## Architecture
|
|
|
|
```
|
|
┌──────────┐ HTTP :3003 ┌──────────────┐
|
|
│ Browser │──────────────▶│ OpenCode │
|
|
│ │◀──────────────│ (host daemon)│
|
|
└──────────┘ └──────┬───────┘
|
|
│ LLM inference
|
|
▼
|
|
┌──────────────┐
|
|
│ llama-server │
|
|
│ (Docker) │
|
|
└──────────────┘
|
|
```
|
|
|
|
## Files
|
|
|
|
- `manifest.yaml` — Service metadata (port, health endpoint, GPU backends, feature definition)
|
|
|
|
## Troubleshooting
|
|
|
|
**OpenCode not accessible on port 3003:**
|
|
|
|
Since OpenCode is a host systemd service, check its status with systemd:
|
|
```bash
|
|
# Check service status
|
|
systemctl status opencode
|
|
|
|
# View logs
|
|
journalctl -u opencode --follow
|
|
|
|
# Restart the service
|
|
systemctl restart opencode
|
|
```
|
|
|
|
**LLM inference not working:**
|
|
- Verify llama-server is running: `docker compose -f ods/docker-compose.base.yml ps llama-server`
|
|
- Ensure sufficient VRAM (minimum 8 GB) is available
|
|
|
|
**Port 3003 already in use:**
|
|
- Check what is using the port: `lsof -i :3003`
|
|
- Update the port in your ODS configuration and restart the service
|