Open Webui Skill

open-webui-skill

Purpose

Use this skill to operate Open WebUI with the bot user configured by OPEN_WEBUI_API_KEY.

Environment

Required environment variables:

• OPEN_WEBUI_BASE_URL: Open WebUI base URL, for example http://open-webui:8080.
• OPEN_WEBUI_API_KEY: bot user API key. Never include this value in prompts, output, logs, reports, or scheduled job definitions.

Optional environment variables:

• OPEN_WEBUI_DEFAULT_CHANNEL: fallback channel when a command omits --channel.
• OPEN_WEBUI_PUBLIC_URL: browser-facing Open WebUI URL used when reports need clickable links.
• OPEN_WEBUI_CHANNELS_LIST_PATH: channel listing endpoint path.
• OPEN_WEBUI_CHANNELS_POST_PATH_TEMPLATE: post endpoint path template containing {channel_id}.
• OPEN_WEBUI_CHATS_ALL_DB_PATH: admin endpoint for all-user chat export. Default is /api/v1/chats/all/db.
• OPEN_WEBUI_MODELS_PATH: model listing endpoint path.
• OPEN_WEBUI_MAX_MESSAGE_CHARS: maximum Markdown post size before truncation.
• OPEN_WEBUI_TIMEOUT_SECONDS: HTTP timeout in seconds.
• OPEN_WEBUI_MAX_RETRIES: retry count for transient API failures.

For API-key setup, endpoint allowlisting, exact routes, and request/response details, read REFERENCE.md.

Capabilities:

• List models visible to the bot user.
• Fetch, create, update, and append to chats visible to the bot user.
• Fetch all users' chats through the admin export endpoint when the API key has admin access.
• Fetch knowledge bases visible to the bot user.
• List channels visible to the bot user.
• Resolve #channel-name or a plain channel name to a channel ID.
• Fetch channel metadata.
• Fetch channel messages, regardless of whether they were written by AI or humans.
• Fetch message threads.
• Fetch file metadata and extracted file content.
• Post Markdown messages to a channel.

Commands

Models:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py models list
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py models get --model llama3.2

Chats:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py chats list --all
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py chats list --all-users
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py chats get --chat-id "<chat_id>"
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py chats post \
  --chat-id "<chat_id>" \
  --model llama3.2 \
  --content "追加メッセージ"

Knowledge:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py knowledge list --details
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py knowledge get --knowledge-id "<knowledge_id>"

Channels:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py channels list
python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py channels resolve --channel report

Fetch channel messages:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py channels messages \
  --channel report \
  --all \
  --include-threads

Fetch file content:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py channels file-content --file-id "<file_id>"

Post to Channels:

python3 /opt/inferlab/skills/open-webui-skill/scripts/client.py channels post \
  --channel report \
  --content "## Report\n\nBody"

Rules

• Never include OPEN_WEBUI_API_KEY in prompts, output, or saved cron definitions.
• Read REFERENCE.md before changing API paths, adding new Open WebUI endpoints, or diagnosing API-key endpoint restrictions.
• Do not use --dry-run for scheduled posting unless the user explicitly asks for preview only.
• For live posting, inspect command JSON. "ok": true means the API accepted the post. "ok": false means no post happened.
• Prefer explicit channel names or IDs. report and #report are both accepted.
• If file content is needed, first inspect message data.files[*].id, then call file-content.

Composition

• For activity reports: fetch messages with this skill, summarize with llm-activity-report-skill, then post with this skill.
• For technology news: collect sources with RSS/search, format with tech-news-report-skill, then post with this skill.