Traditional math AI tools follow: user asks → AI freely generates answer. This relies on AI generation ability, prone to hallucination, cannot guarantee mathematical precision.

Dmmsys Math Knowledge Client reverses this: user asks → query deterministic knowledge base first → return exact result on hit (no AI) → fall back to AI reasoning only on miss. This guarantees precise answers with zero hallucination for knowledge-base-covered questions.

1)    Knowledge base hit priority: questions found in the knowledge base return structured theorems, proofs, formulas directly, no AI generation.

2)    Audited fallback reasoning: when knowledge base misses, AI reasoning is used, with audited results enterable as new deterministic knowledge.

3)    Result source labeling: each query result labels its source (knowledge base hit / AI reasoning / other), users know result reliability.

4)    Structured return: results include theorem name, description, LaTeX formula, proof steps, validation status, dependencies.

5)    Zero-hallucination design: never fabricates within knowledge base coverage, deterministic queries guarantee mathematical precision.

Math Knowledge Client uses a lightweight triad architecture: Client = Agent, Local Database = CSV, Console = HTML. Zero third-party dependencies, pure standard library, standalone execution.

This means zero-install deployment - just double-click to run. CSV is human-readable, Git-friendly, openable in Excel, directly editable in text editors, fully auditable.

1)    Client = Agent: standalone process provides API service handling knowledge queries, sentence datification, solve records.

2)    Local Database = CSV: query records, datification results, solve records all stored as CSV, human-readable, Git-friendly.

3)    Console = HTML: single-file HTML console, three-column layout, real-time formula rendering, no UI framework needed.

4)    Zero third-party dependencies: pure Python standard library, no external packages needed, standalone execution.

5)    Packable distribution: supports packaging as standalone executable and installer with desktop shortcut.

Math Knowledge Client provides not only knowledge queries but also sentence datification: decomposing natural language text sentence-by-sentence into components, sentence patterns, semantic arguments, and lexical tokens.

Formula rendering supports real-time LaTeX rendering with built-in common formula snippets. Server-side pre-rendering falls back to CDN on failure, ensuring formulas always display.

1)    Sentence datification: decomposes Chinese, English, math symbol text into precise data objects, the entry point for semantic-to-logic-to-code engine.

2)    Multi-language support: supports Chinese, English, math symbols and other languages for text datification.

3)    Real-time LaTeX rendering: input LaTeX instantly renders to math formulas with built-in snippet library.

4)    Server-side pre-rendering: server pre-renders formula HTML, falls back to CDN on failure, ensuring rendering success.

5)    Three-column console: formula input/render column, natural language to math system column, local memory column.

Math Knowledge Client is an offline client: local CSV is primary storage, works normally without network. When online, local increments sync to server in both directions.

This means the client works regardless of network conditions. Offline operations use local data, online auto-syncs, no data loss.

1)    Offline capable: recording, listing, statistics all read local CSV, zero network dependency, works without network.

2)    Upload sync: uploads unsynced local solve records to server, marks as synced on success.

3)    Download sync: pulls problems from server, adds to local ledger if missing, stores only readable text not server internals.

4)    Dedup reconciliation: compares local and server records to avoid duplicate downloads.

5)    Graceful degradation: sync interface returns offline status when disconnected, no crashes, no data loss, auto-resumes on reconnection.

Math Knowledge Client integrates WeChat Open Platform website application QR login, with login state stored in local CSV, no external database dependency.

Users log in by scanning QR code with WeChat, no separate registration needed. After login, personal solve records and learning progress can be associated.

1)    Standard OAuth2 flow: frontend scan → WeChat authorize → callback token exchange → session established, complete OAuth2 snsapi_login.

2)    Local session management: login state stored in local CSV, no external database, verifiable offline.

3)    Graceful degradation: unconfigured WeChat app shows degraded prompt, does not block other features.

4)    Environment variable configuration: all WeChat config via environment variables, no hardcoding, secure and controllable.

The client provides complete solve record management: offline recording, local statistics, online sync. Also supports problem management: create, list, sync, delete.

Solve records include timestamp, problem code, name, type, difficulty, resolved status, solution steps, duration, notes - supporting category statistics and accuracy analysis.

1)    Complete record fields: timestamp, code, name, type, difficulty, is_resolved, steps, duration, note, synced, source.

2)    Plain text steps: steps stored as plain text list with delimiters, not JSON, following plain-text ledger conventions.

3)    Local statistics: total count, solved count, category accuracy, unsynced count all from local CSV, offline capable.

4)    Problem management: create custom problems, list queries, sync to server, delete - complete CRUD operations.

5)    Bidirectional sync report: shows upload success/fail counts, download counts after sync, transparent sync status.

Math Knowledge Client provides an OpenAI-compatible interface with SSE streaming output. Any AI tool supporting OpenAI interface can directly connect.

The client wraps math knowledge query capability as standard AI conversation interface, AI tools can call directly to get precise math knowledge answers rather than free generation.

1)    OpenAI compatible: provides standard chat/completions interface, supports messages format and SSE streaming output.

2)    DSH direct connection: AI Harness tools can directly configure baseURL, no extra adaptation needed.

3)    Hybrid mode metadata: streaming output includes intent recognition, proof nodes, validation status metadata for frontend display.

4)    AI-assisted solving: optionally forwards to user-configured AI API for reasoning, defaults to frontend direct connection for key security.

5)    Universal command dispatch: provides universal command interface supporting multiple business commands.