AI Wisebase

Creates an asynchronous video explanation task for a given concept. Sends a request to an external video generation service and returns a taskId to track progress. No existing data is modified or deleted.

Internal APP fetch tool for accessing internal APIs

Generates flash cards based on the user's input, Each flash card should contain a question and an answer.

Generate a non-hierarchical, network-based knowledge graph visualization.

Generate a quiz with multiple-choice questions on a given topic. You (ChatGPT) should generate the quiz content including questions, hints, options with explanations for each option, then pass the complete data to this tool for display.

IMPORTANT: For mathematical formulas, use KaTeX syntax:

• Inline formulas: $formula$ (e.g., $x^2 + y^2 = z^2$)
• Display (block) formulas: $$formula$$ (e.g., $$\frac{-b \pm \sqrt{b^2-4ac}}{2a}$$)

KaTeX syntax examples:

• Fractions: $\frac{a}{b}$
• Powers: $x^2$, $e^{i\pi}$
• Subscripts: $a_n$
• Square root: $\sqrt{x}$
• Greek letters: $\alpha$, $\beta$, $\pi$

Retrieves the current status and result of a video explanation task by its ID. This is a read-only operation; no data is created, modified, or deleted.

Opens the Wisebase file browser panel, displaying the user's uploaded files and knowledge base documents. This tool only renders the file list widget; no data is created, modified, or deleted.

display use's flash cards list When to use:

• User asks: "Show me my saved flash cards"
• User asks: "Display my flash cards"
• User asks: "Load the flash cards I saved"
• User asks: "Show my flash cards history"

you can call the tool

display use's knowledge graph list.

display use's quiz list When to use:

• User asks: "Show me my saved quiz"
• User asks: "Display my quiz"
• User asks: "Load the quiz I saved"
• User asks: "Show my quiz history"

you can call the tool

When a user wants to create a document that they intend to save for future use, this tool should be invoked to analyze and generate the document according to the user's instructions; after the user confirms, the save function will then be called. Otherwise, the save function should not be invoked without the user's permission.

The model must not actively invoke this tool during the conversation. This tool is only intended to be triggered by the frontend via window.openai.calltool after chat-save-preparation has been called, and it is used solely for saving files that have already been edited.

Therefore:

• The model must not call this tool directly in its reasoning or responses.
• The timing and logic for invoking this tool are entirely controlled by the frontend, and are independent of any tool calls produced by the model.

• Task: Agentic RAG with multi-round iterative retrieval, up to 20 rounds. Not a single-call tool.
• Protocol:
• One short query per round: use noun phrases or minimal short questions; avoid long sentences.
• Call retrieval → read passages and should_continue.
• If should_continue = true: you must continue, and the new query must be substantially different from history; if false: stop and answer.
• Query diversity requirements (avoid duplicate hits):
• Strictly forbid synonym rewrites or morphological variants: user book acquisition ≠ user acquiring books (counted as duplicate).
• Switch dimensions:
• Question type: what/how/why/when/who
• Granularity: overview → specifics → edge cases/counterexamples
• Perspective: user/system/business
• Domain: features/behavior/architecture/performance/algorithms/data/security/cost
• Complementary info: if you got “benefits”, next ask about “challenges/implementation/constraints/tradeoffs”.
• Required micro-analysis before each new query (internal only, do not output to user):

1) Summarize key points from current passages in 1–2 lines 2) List missing angles (1–3 items) 3) Pick one missing angle 4) Generate a brand-new keyword set targeting only that angle, with different terms

• Query generation rules:
• Keywords only, separated by commas or spaces; no filler words or auxiliaries; no long qualifiers.
• Prefer domain nouns and technical terms; optionally add steering terms like tradeoffs, limitations, metrics, workflow.
• Each round must replace prior core stems or switch to a new dimension/perspective/domain.
• Example (short queries, multiple angles):
• R1: user book acquisition
• R2: PDF reader personalization features
• R3: reading behavior patterns, statistics
• R4: knowledge base architecture, retrieval
• If R4 returns should_continue = false, then stop and answer.
• Remember:
• Same query → same results → wasted call
• should_continue = true → immediately query again from a different angle
• More than 5 rounds are allowed, continue until information is sufficient.

When a user indicates that they want to translate a file—especially a PDF—call this tool. The user does not need to have uploaded the file beforehand; the tool will guide them through uploading the file after it is invoked.

Opens the Wisebase file upload interface. This tool only renders the upload panel for user interaction; no backend data creation, modification, deletion, or file transfer is performed by the tool itself.