opencode-analyze-image

A cybernetic eye for text-only models: an external vision model that lets them see images.

简体中文 · Report an issue

"The reason is that this, most of all the senses, makes us know and brings to light many differences between things."

— Aristotle, Metaphysics, Book I, Part 1

This is an independent community plugin for OpenCode and is not built by or affiliated with the OpenCode team.

In Action

_{Image analysis running inside an OpenCode terminal session.}

Installation

Prebuilt JavaScript file

Download analyze_image.js from the matching GitHub Release:

mkdir -p ~/.config/opencode/plugins
curl -fL https://github.com/cipherTing/opencode-analyze-image/releases/download/v0.1.2/analyze_image.js \
  -o ~/.config/opencode/plugins/analyze_image.js

OpenCode automatically loads JavaScript and TypeScript files from ~/.config/opencode/plugins/ at startup. Project-level local plugins use .opencode/plugins/ instead.

When OPENCODE_CONFIG_DIR is set, use that directory instead of ~/.config/opencode.

After installation, restart OpenCode and use it normally.

Optional terminal status

The terminal status is an optional TUI surface. The image-analysis tool works without it. To show · Vision ON below the terminal input, configure the TUI entry separately.

For the prebuilt JavaScript installation, download the TUI file and add its path to ~/.config/opencode/tui.json:

mkdir -p ~/.config/opencode/tui
curl -fL https://github.com/cipherTing/opencode-analyze-image/releases/download/v0.1.2/analyze_image_tui.js \
  -o ~/.config/opencode/tui/analyze_image_tui.js

{
  "plugin": [
    "./tui/analyze_image_tui.js"
  ]
}

Configuration

Create the configuration file at:

~/.config/opencode/analyze_image/config.json

When OPENCODE_CONFIG_DIR is set, use $OPENCODE_CONFIG_DIR/analyze_image/config.json instead.

Minimal configuration

{
  "trigger_models": [
    "deepseek/deepseek-v4-flash"
  ],
  "api_format": "openai_chat",
  "base_url": "https://api.openai.com/v1",
  "model": "your-vision-model",
  "api_key": "your-api-key",
  "reasoning": {
    "effort": "medium",
    "adaptive": true
  }
}

Required fields

Field	Description
trigger_models	Primary models that can use the plugin. Use the full provider/model identifier.
api_format	openai_chat, openai_responses, or anthropic_messages.
base_url	Auxiliary vision model API endpoint.
model	Auxiliary vision model ID.
api_key	Vision model API key, or an {env:NAME} / {file:PATH} reference.

Optional fields

Field	Default	Description
reasoning.effort	medium	Allowed values: none, low, medium, high, xhigh, or max.
reasoning.adaptive	true	Anthropic Messages only: true uses adaptive thinking; false uses budget-based thinking.
timeout_seconds	120	Request timeout.
max_retries	2	Number of request retries.
max_output_tokens	4096	Maximum visible output length.

API keys

The key may be written directly, read from an environment variable, or read from a file:

"api_key": "your-api-key"

"api_key": "{env:ANALYZE_IMAGE_API_KEY}"

"api_key": "{file:~/.secrets/analyze-image-key}"

Relative file paths are resolved relative to config.json.

trigger_models uses exact matching. Keep the complete identifier when a model ID contains additional slashes.

Image size and directory limits can use the defaults in config.example.json. Change the image section only when needed.

Full Configuration

See config.example.json for the complete field list.

License

MIT