Built-In Tool Reference for Prompt Customization

If you customize TomoriBot’s system prompt, persona instructions, or external provider prompt templates, prefer the stable prompt macros below instead of hardcoding tool names.

• Prompt macros like {memory_tool} are expanded during context assembly. Exact tool names are emitted wrapped in backticks, while unresolved search/fetch families fall back to plain-language text.
• Base Tool means the tool is part of TomoriBot’s normal built-in tool set. It may still depend on the current provider/model supporting tool calling.
• Other requirements below are additional gates such as server feature flags, Discord permissions, model capabilities, or optional API keys.
• Admin-added MCP tools are intentionally not listed here because their names depend on each server’s configuration.

Built-In Function Tools

Tool name	Prompt macro	Requirements	Purpose
review_capabilities	{capabilities_tool}	Base Tool	Check current chat abilities, slash commands, or runtime settings before answering.
create_long_term_memory	{memory_tool}	self_teaching_enabled	Save a new stable server fact or user-specific preference for future conversations.
update_long_term_memory	{memory_update_tool}	self_teaching_enabled	Replace an outdated long-term memory by ID.
update_short_term_memory	{short_term_memory_tool}	Base Tool; unavailable on NovelAI	Save temporary working memory for the current channel/story arc without making it permanent.
create_task	{task_tool}	Base Tool	Schedule one-time or recurring reminders and self-tasks.
update_task	{task_update_tool}	Base Tool	Replace or delete an existing reminder or self-task by ID:N, scoped to the requester.
cross_channel_message	{cross_channel_tool}	Base Tool; unavailable on NovelAI; target channel permissions and cross-channel blocklist still apply	Instantly act in another channel or thread, with optional boomerang report-back.
create_thread	{create_thread_tool}	thread_creation_enabled; bot CreatePublicThreads and SendMessagesInThreads permissions	Create a public thread in the current or named channel and send its starter message.
select_sticker_for_response	{sticker_tool}	sticker_usage_enabled; USE_EXTERNAL_STICKERS	Pick a matching server sticker to accompany the response.
manage_message	{manage_message_tool}	manage_message_enabled; MANAGE_MESSAGES still required for pin	Pin any recent message, or edit/delete recent messages sent by Tomori or its characters.
block_user	{block_user_tool}	user_blocking_enabled; active server persona context	Temporarily mute or block a Discord user for the active persona only. mute prevents the target from triggering that persona; block also hides that user’s recent live dialogue-history messages/media from that persona’s context. This does not forget or redact memories.
unblock_user	{unblock_user_tool}	user_blocking_enabled; active server persona context	Remove an active persona-scoped mute/block for a Discord user.
interact_with_recent_message	{message_interaction_tool}	Base Tool; normal Discord send/react capability still applies at runtime	React to a recent message or send a short backtracking reply to it.
peek_profile_picture	{profile_picture_tool}	Base Tool; requires either a vision-capable chat model or a configured vision_llm	Inspect a user’s avatar or the active persona avatar.
read_document	{document_tool}	Base Tool	Extract text from a PDF, TXT, or MD attachment in a recent message.
reveal_message_metadata	{message_metadata_tool}	Base Tool	Annotate recent visible turns with ref_N handles and sent timestamps for precise message targeting.
increase_media_context	{media_context_tool}	Base Tool; requires a vision-capable chat model	Pull older hidden images/videos back into context when media was windowed out for optimization.
process_gif	{gif_tool}	Base Tool; development only; requires a vision-capable chat model	Extract keyframes from a GIF for analysis.
process_youtube_video	{youtube_tool}	Base Tool; requires a model with YouTube/video support	Analyze a specific YouTube link on demand.
analyze_image	{image_analysis_tool}	Base Tool; requires a configured vision_llm; only shown when the current chat model cannot already see images	Delegate image understanding to a separate vision model.
generate_image	{image_generation_tool}	imagegen_enabled; active provider must support native image generation	Generate an image with the current provider; uses Physical Appearance context, default positive image tags, and default negative tags when the backend declares negative-prompt support; reference, inpaint, and outpaint arguments are shown only when the configured backend supports them. target_identity accepts one or more user/persona names, each resolved to that user’s guild avatar (or alter persona webhook avatar) and added as a reference. The final Discord message uses a Components V2 media gallery with generation-time subtext, including a line listing any referenced users/personas.
generate_image_nai	{anime_image_generation_tool}	imagegen_enabled; NovelAI provider or NovelAI optional API key	Generate or edit anime-styled images with NovelAI, using Physical Appearance context as character tags when relevant. The final Discord message uses a Components V2 media gallery with generation-time subtext.
generate_voice_message	{voice_message_tool}	ElevenLabs optional API key; active persona needs an ElevenLabs voice; voice_message_enabled	Send a spoken Discord voice reply instead of plain text.

Default Search / Web Extras

These are the common built-in or bundled web tools Tomori can expose when web access is enabled. Exact availability depends on provider support, server config, API keys, and which MCP servers are active.

Family macros below may resolve to the listed bundled tools or to compatible guild MCP replacements when admins register their own web_search or url_fetcher servers.

Unified web_search

A single LLM-visible tool replaces the previous four brave_* tools. Its assembled category enum is narrowed for the active backend before the provider adapter sees it. SearXNG exposes all categories (text / image / video / news / science / it / files / music), Brave exposes the common four categories (text, image, video, news), and DDG/Felo expose text only. If no backend is available, web_search is omitted.

Search progress embeds show the selected category as a separate localized label (for example, “Searching videos for beaver…”) rather than mutating the query string.

Tool name	Prompt macro	Requirements	Purpose
web_search	{web_search_tool} / {image_search_tool} / {video_search_tool} / {news_search_tool}	web_search_enabled; at least one search backend currently available	Search the web. The assembled category arg only lists categories supported by the active backend. Optional count arg sets result count (image: max 10 sent to Discord; text-like categories: max 20 in result list). The dispatcher hides engine selection from the model — saves ~400 tokens/turn vs. the previous 4-tool surface.
fetch_url	{url_fetch_tool}	web_search_enabled; active bundled fetch path; unavailable on NovelAI	Read a specific web page or URL in more detail. Arguments mirror the bundled MCP fetch server: url, optional max_length, optional start_index, optional raw.
url-metadata	{url_metadata_tool}	web_search_enabled; active DuckDuckGo/Felo MCP search server	Retrieve page metadata for a URL when a metadata-specific fetcher is available.

Engine-internal details: the Brave per-category implementations live under src/tools/restAPIs/brave/internal/ as InternalBrave* services consumed by webSearch/braveEngine.ts. They are intentionally not LLM-visible. The DDG/Felo paths are reached through webSearch/duckduckgoEngine.ts / feloEngine.ts, which call the MCP server directly via DuckDuckGoHandler.executeWebSearchInternal() / executeFeloSearchInternal(). SearXNG is reached through webSearch/searxngEngine.ts, which calls the self-hosted /search endpoint via restAPIs/searxng/. Adding a new engine = implement WebSearchEngine and append to the chain in webSearch/dispatcher.ts.

Unified fetch_url

fetch_url is the single bundled URL-reading tool shown to the LLM. Its dispatcher can try optional browser sidecars first, then always falls back to the internal mcp_fetch engine. Before dispatch, TomoriBot blocks localhost/private/internal/reserved target URLs unless FETCH_URL_ALLOW_PRIVATE_NETWORK=true; the default error message explicitly names the FETCH_URL_ALLOW_PRIVATE_NETWORK=false default so the bot can explain the failure to the user. mcp_fetch calls the existing bundled MCP fetch server and reuses its result processing. The raw global MCP function name fetch is hidden from the LLM after centralized feature-flag filtering.

Guild MCP replacements still work: if an enabled guild MCP server is registered as url_fetcher and exposes functions, TomoriBot hides the bundled fetch_url for that guild so {url_fetch_tool} resolves to the guild function instead.

Crawl4AI sidecar (optional URL-fetch engine)

Crawl4AI is an optional browser-rendered markdown sidecar used only behind fetch_url. When CRAWL4AI_BASE_URL is set and the health probe passes, it runs before mcp_fetch in the engine chain. If unset or unavailable, fetch_url falls back to mcp_fetch transparently.

See Setup: Crawl4AI Sidecar for deployment, cookie injection, and FETCH_URL_ENGINE_ORDER configuration.

SearXNG sidecar (optional self-hosted engine)

SearXNG is a privacy-respecting metasearch aggregator that sidesteps single-engine rate limits and unlocks SearXNG-only search categories: science, it, files, and music. It sits between Brave and DDG/Felo in the engine chain. If unset or unavailable, the chain reduces to Brave → DDG → Felo.

See Setup: SearXNG Web Search Sidecar for deployment, image result tuning, and AWS/GCP configuration.