Agent Tools Reference
Nephele's Agent is a tool-calling loop: understand what you need → pick the right tool → execute → keep reasoning on the result, until the task is done. The current version registers about 66 tools, plus two cloud-only tools, cloud_search and delegate_tasks, once you enter Cloud MAX mode. Tools load dynamically on demand—only the core set ships initially, and tool_search pulls in the rest when you need them.
You don't need to memorize any of this. The Agent picks tools automatically from your plain-language description. This reference is for advanced users who want to understand how Nephele works under the hood, or who reference tool names when writing a Skill.
Search & References
| Tool | Purpose | Key parameters |
|---|
find_references | Search reference images across platforms—covering Pixiv / Konachan / Pinterest / Huaban / ArtStation / Wikimedia / local Eagle | keywords search terms, site specific site, sources source filter (defaults to Eagle only; online sources must be enabled explicitly) |
eagle_search | Search assets inside your Eagle library | keyword, ext file extension, tags tag array, star rating, folders, limit count |
find_similar_by_image | Find visually similar images (local Eagle, EVA02 tagger embeddings) | image_path local image path, eagle_id Eagle asset ID (one of the two), top_k number returned |
find_by_text | Text-to-image: a single English description finds the closest matching references in Eagle (CLIP text-image space) | query English description (required), top_k number returned (default 30, max 60) |
search_files | Search local files by Glob pattern | pattern match pattern, path search directory, max_results cap, offset pagination offset |
reverse_image_search | Reverse image search (trace where an image came from) | image_path image path, mode mode (default/resolve) |
resolve_source_url | Resolve an image's original source URL | url the link to resolve |
show_reference_picker | Pop up the reference picker (for you to choose and pin images) | source_page source page URL (optional) |
Notes:
find_references already searches multiple platforms in parallel internally—do not use delegate_tasks to split it into multiple subtasks. It searches local Eagle by default; online sources (Pixiv, etc.) must be enabled explicitly in sources.show_reference_picker is usually called after browser_extract(type="images"); the image list is read automatically from the browser cache.
Information Retrieval
| Tool | Purpose | Key parameters |
|---|
web_fetch | Fetch web page content and convert it to Markdown | url target address, extract_images whether to extract page images |
cloud_search | Cloud time-sensitive search (news / prices / recent events, Cloud MAX mode) | query keywords |
download_url | Download a file from a URL into the local cache | url download address, filename optional filename |
Notes:
web_fetch has a 15-minute cache and automatically retries on SSL errors. It is the Agent's main way of pulling data from the web.cloud_search is only available in Cloud MAX mode and calls the Nephele API's grounding search.
File System
| Tool | Purpose | Key parameters |
|---|
read_file | Read file content | path file path, offset start line, limit max lines |
write_file | Write a file | path, content, append whether to append |
edit_file | Edit a file (find and replace) | path, old_text original, new_text replacement, replace_all whether to replace all |
list_dir | List directory contents | path directory path, show_hidden whether to show hidden files |
grep_content | Search file contents for text/regex | pattern search content, path directory, glob_filter file filter, max_results cap, context_lines lines of context |
Safety limits: All file operations are bound by an allowlist and cannot access sensitive system directories. grep_content automatically skips binary files and files larger than 1MB.
Eagle Library Management
| Tool | Purpose | Key parameters |
|---|
eagle_import | Import files into Eagle | path file path, name, tags tag array, folder_id target folder, annotation note |
eagle_library_info | Get Eagle library info | library_path library path, action operation type |
eagle_update | Update asset metadata (tags / rating / note / move folder) | ids asset ID array (or item_id for a single one), tags, tag_mode mode (add/remove/replace), star rating, annotation note, folder_id |
eagle_folder_create | Create an Eagle folder | name folder name, parent_id parent folder ID (optional) |
eagle_artist_trace | Batch-trace the artist behind Eagle assets, using zero-cost signals (source URL / filename) to write back an "Artist:" tag | folder_id, keyword, limit count, apply whether to write back (default true), max_network network cap (default 300) |
Browser Automation
| Tool | Purpose | Key parameters |
|---|
browser_navigate | Navigate to a given URL | url, timeout timeout in seconds |
browser_action | Perform a browser action | action (click/fill/scroll/hover/screenshot), ref_id element identifier, text fill text, direction scroll direction |
browser_snapshot | Capture a snapshot of the current page | interactive_only interactive elements only, annotated whether to annotate |
browser_extract | Extract page content | type (images/links/text), min_size minimum image size |
AI Evaluation & Analysis
| Tool | Purpose | Key parameters |
|---|
evaluate_image | AI multi-dimensional image quality evaluation | image_path, custom_prompt custom evaluation dimensions, style_description, work_category |
analyze_style | Analyze style DNA (StyleDNA) | image_path |
detect_ai | Detect whether an image is AI-generated | image_path |
Artist & Work Exploration (Cloud MAX)
This group helps you branch outward from one artist or one piece to explore inspiration. Most are online and Cloud MAX–only.
| Tool | Purpose | Key parameters |
|---|
find_artist_works | Pull about 12 recent works from an artist for study (Pixiv first, X as fallback), with enforced source attribution and no redistribution | artist_name, pixiv_user_id, twitter_handle (all optional) |
find_similar_artists | Discover 12 stylistically similar artists via Pixiv's recommendation graph | artist or illust_url, limit (default 12, max 20) |
find_similar_works | Given a Pixiv piece, return 12–18 works similar in style/composition (across artists) | illust_url (required), limit (default 12, max 18) |
wiki_art_context | Use Wikidata to trace art-history context: movement, influences, key works, lifespan—or work backward from a movement to its representative artists | query or wikidata_id |
explore_related_concepts | Use a Wikipedia summary + related concepts to brainstorm, producing new search / moodboard seeds | concept (required) |
launch_wisp_browser | When find_references reports Wisp is not connected, launch Edge and wait for the extension to be ready | wait_seconds (default 10, max 20) |
Reference Boards & Moodboards
Let the Agent read and write the pinned reference board directly, and assemble moodboards automatically.
| Tool | Purpose | Key parameters |
|---|
get_board_state | Read the current reference board: each item's id, type, and visible bounding box | none |
annotate_board | Append annotations to the reference board: arrows / lines / rectangles / ellipses, text cards, or composition overlays (rule of thirds, etc., requires targetId) | annotations annotation array |
build_moodboard | Assemble a set of references into a moodboard: visual hierarchy + extracted color swatches + caption, drop it onto the reference board and export a composited PNG | eagle_ids[], image_paths[], caption, intent, board_name, target_count (default 12) |
Creator Data & Commission Market
| Tool | Purpose | Key parameters |
|---|
get_dashboard_stats | Query the creator data center: library (Eagle library profile) or snapshot/history/works (per-platform performance) | action (library/snapshot/history/works), platform, metric_key, days, sort_key, limit |
mihuashi_fetch | Pull public Mihuashi data (read-only): commission projects / details / market rates / showcase | action (projects/detail/market/showcase), query, project_id, page, max_pages (max 5) |
Publishing & Packaging
| Tool | Purpose | Key parameters |
|---|
pack_image | Package delivery files (watermark / crop / format conversion) | input_path, watermark_path watermark path, output_dir, watermark_mode watermark mode |
generate_publish_package | Generate a multi-platform publish package | image_path, platforms target platform array |
generate_copy | Generate social media copy | image_path, custom_prompt custom prompt |
upload_web | Upload a Godot Web export to play.arisfusion.com | export_dir export directory (containing index.html), project_name project name |
Project Processing
| Tool | Purpose | Key parameters |
|---|
clean_psd | Clean a PSD file (remove empty / hidden layers) | file_path, remove_hidden delete hidden layers, remove_empty delete empty layers |
export_godot | Export a Godot project as exe or Web | project_path project directory, format (exe/web), output_path (optional) |
Provenance & Timestamping
| Tool | Purpose | Key parameters |
|---|
timestamp_files | Generate trusted timestamps for files | file_paths file path array, output_dir |
verify_timestamp | Verify a timestamp certificate | file_path certificate path |
protect_works | Rights-enforcement protection workflow | file_paths work path array, author_name, output_dir |
System & General
| Tool | Purpose | Key parameters |
|---|
run_python | Execute Python code | code code string, script_path script path (one of the two), description |
open_app | Open an external application | name application name |
open_url | Open a web page in the system browser | url |
open_path | Open a folder or file | path file or folder path |
UI Control (Internal)
This group lets the Agent operate Nephele's own interface directly, and you can watch the interface change on screen in real time.
| Tool | Purpose | Key parameters |
|---|
ui_navigate | Switch the interface view | view view name |
ui_set_field | Set a UI element property | object_name element name, property, value |
ui_click | Click a UI button | object_name element name |
ui_read_state | Read UI state | object_name element name, property |
ui_wait_for | Wait for a UI state change | object_name element name, property, expected expected value, timeout (seconds) |
navigate_to_tool | Open a built-in tool page | tool_name (ai_detect/ai_evaluate/packer/eagletagger), file_path preloaded file (optional) |
picker_context | Read a condensed state of the reference picker (source, artist, Eagle id, and path of the first / selected / preview images); prefer it when handling follow-up picker operations | action (summary/item/selected/preview/list), target, index, limit, offset |
request_demo | Trigger an in-app step-by-step demo / tutorial (pilot builds only) | tool_name demo name (e.g. packer_demo / first_run) |
Task Management & Coordination
| Tool | Purpose | Key parameters |
|---|
delegate_tasks | Run multiple independent subtasks in parallel (Cloud MAX only) | tasks task array (each with a description and an optional tools allowlist) |
progress_update | Report progress on a multi-step task | steps step array (each with a name and a status), title task title |
ask_user_question | Pause the loop and pop up 1–4 choice questions for you to decide | questions question array (each with question, header, options, multiSelect) |
tool_search | Dynamically load the full schema of an inactive tool | query (select:tool_name for exact load, or keyword search), max_results count |
Notes:
delegate_tasks is a Cloud MAX–only tool, supports up to 6 parallel subtasks with a 120-second timeout each, and runs them truly in parallel.tool_search is Nephele's lazy-loading mechanism: only core tools are loaded initially, and when needed this tool dynamically loads the full parameter definitions of other tools to save tokens.
Memory System
| Tool | Purpose | Key parameters |
|---|
memory_write | Write a persistent cross-session memory (same name overwrites) | name, type (user/workflow/project/reference/correction), description, body |
memory_load | Read the full content of a memory by name | name |
memory_delete | Delete a memory by name | name |
Permission Levels
Every tool carries a permission label, and the Agent validates it before calling:
| Permission | Description | Representative tools |
|---|
READ | Read-only file system | read_file, list_dir, grep_content |
WRITE | Read and write files | write_file, edit_file |
NETWORK | Network access | web_fetch, download_url, cloud_search, upload_web |
EXECUTE | Run external programs | run_python, open_app, export_godot |
Tool Selection Tips
Finding references: Use find_references directly—don't break up your phrasing. It already searches multiple platforms in parallel internally (Eagle only by default; enable online sources in sources).
Tracing an image's origin: reverse_image_search to trace the source → resolve_source_url to resolve the original page.
Organizing Eagle: eagle_search to find the targets → eagle_update to batch-edit tags/ratings/move folders by ID; use eagle_artist_trace to batch backfill artist tags.
Writing code/scripts: run_python is the Agent's "Swiss army knife"—it can run any Python code to process data, convert formats, batch-rename, and more.
Browser workflow: browser_navigate → browser_extract → show_reference_picker is the standard chain for collecting reference images in bulk from a web page.
Last updated Jun 21, 2026·Applies to v0.5.2-beta