New here? Open the User Manual (PDF) — a plain-English, picture-filled guide to every feature.

The KDP Niche Intelligence Dashboard is a self-contained, offline-first research tool for Kindle Direct Publishing (KDP). It evaluates 130 curated low-content / journal niches against the scoring rubric of Mastering KDP in 2025, scores each niche out of 14, decides a verdict (PASS / CONDITIONAL / AVOID), and turns every decision into an actionable publishing roadmap. Everything runs locally in pure Python — no pip packages, no frameworks, no tracking. The niche engine, database, and playbook work 100% offline; only the optional Author Studio (OpenAI-compatible LLM) and Web Search (keyless DuckDuckGo or your own provider) make outbound calls, and only when you trigger them. Phases 1–5 complete: token counting, streaming, chapter editor, HTTPS, multi-user auth with admin panel, encrypted API keys, service worker, and 557 automated tests.

• Niche Library — browse all 130 niches with live search, category, verdict and competition filters; drill into a niche for keywords, title ideas, its full score breakdown, and a one-click live market check with per-niche history.
• Live Market Data — one-click live Amazon checks per niche: search-result counts, average reviews, average price, and a Best Sellers Rank probe of the top result, all scored against the same rubric so a clean market can reach PASS. Choose the source with a persisted toggle — Auto (PA-API when keys are configured, else keyless), PA-API, or Keyless (Amazon autosuggest + search-page scrape). Results are cached for an hour, fetches are throttled and retried with backoff to avoid bot detection, and every check is recorded in the niche's market history.
• Validator — type any niche or phrase. Known niches get their master-report score; brand-new ideas get a provisional score computed on the fly by the same scoring engine.
• Red-Light Scanner — checks a keyword against the global avoid-list (saturated niches, review-farm niches, "Get Rich with YouTube" niches, generic keywords) and explains the flag.
• Playbook — the distilled ruleset: golden rules, BSR target bands, review/search/price rules, royalty examples, content tiers, the global avoid list and a pre-publish checklist.
• My Picks — build a publication pipeline of niches you plan to write.
• Pipeline — the full funnel in one place. Pick a validated niche (or type a new one) and click Run Full Pipeline: it validates against the live market, creates the book draft with its keywords and title ideas, and generates a 4-phase launch & marketing plan — then tracks every pick, its books and its launch progress on one board.
• Launch Plan Generator — one button inside a book turns the niche into a concrete 4-phase selling plan (Pre-Launch: title/cover/description keywords, categories, ARC reviews · Launch: pricing, KDP Select free runs, review velocity · Post-Launch: rank defence, listing tweaks · Scale: ads, ACOS, series). Generated by the configured LLM and grounded on live market data; a deterministic KDP-grounded plan is used offline.
• Memory — store notes, rules, decisions and PDFs (including the master validation report) for future reference.
• History — audit trail of every validation you run, with verdict and score.
• Author Studio — plan a book from any niche or topic, write it chapter by chapter with an inline editor, copy-edit it, audit the draft with local quality metrics, plan a series, track the launch pipeline, and export a KDP-ready manuscript as EPUB / DOCX / PDF / HTML / MD / TXT. Uses the OpenAI API when a key is configured; otherwise runs in offline demo mode. Chapters can be streamed via SSE in real time.
• Book Assistant — an embedded chat assistant with the dashboard's knowledge baked in (niches, picks, playbook rules and your open book). It can outline, draft, critique, or answer KDP questions. Supports SSE streaming, function-calling (the AI can call web search, market check, and MCP tools itself), and chat history that persists across page refreshes.
• Research Vault — upload source books (TXT / MD / PDF, text extracted locally) or fetch web pages into a private corpus; auto-analysis on upload; "Ask the Vault" Q&A with numbered citations; "Authoring Brief" for multi-source briefs; and "Ground with research" so chapters and assistant answers are built only from your vetted material.
• Deep Web Search — one query across every source you enable in Services → Deep Search Sources (keyless DuckDuckGo, DuckDuckGo AI answers, Wikipedia, public SearXNG, free-tier Brave / Tavily / Serper keys, MCP search/fetch servers, OpenAlex academic, Hacker News, Internet Archive, GitHub). It merges and dedupes results, and in Read & brief mode it opens the strongest pages and has the active AI summarize them into a research brief.
• Omni Search — type a word or paste a URL; one query fans out to every deep-search provider, MCP server, web search, the browser-scrape chain, and optional AI synthesis. Results are merged and deduped.
• AI Toolbox — when the active LLM supports function-calling, the Book Assistant can autonomously call any tool (web search, deep research, market check, Amazon suggest, vault search, fetch URL) and any configured MCP server.
• SVG Cover Builder — generate a print-ready book cover as SVG directly in the browser, with customizable title, subtitle, author name, and background colour.
• Diagnosis — a read-only health check of the whole tool with an overall HEALTHY / WATCH / NEEDS-ATTENTION verdict.
• Testing — one-click re-run of the entire automated test suite (557 checks) with a live progress bar.
• Multi-User Auth — PBKDF2 password hashing, secure session cookies, admin panel for user management (create, edit, deactivate, promote). Registration is closed after the first user.
• Undo System — soft-delete with a 30-second undo toast on destructive actions.

Verdicts: PASS (score ≥ 10 and no red lights), CONDITIONAL (needs an angle check before writing), AVOID (saturated, red-flagged, or below threshold).

• Pure Python 3 standard library — no pip packages, no frameworks.
• engine.py — scoring, validation, keyword generation, title ideas, red-light scan, provisional scoring, ROI, roadmap and the playbook ruleset.
• market.py — live Amazon market data: PA-API v5 SigV4 signing, keyless search-page scraping, autosuggest, a Best Sellers Rank probe, short-TTL result caching, and throttled/retrying HTTP with backoff.
• db.py — SQLite persistence (niches, books, research, users, sessions, audit, undo, market cache, migrations).
• ai.py — LLM client + offline demo, book pipeline, token estimation, prompt sanitization, SSE streaming.
• tools.py — AI toolbox: core tools + dynamic MCP tools, ToolRunner for function-calling.
• research.py — text extraction, corpus search, web search, URL fetching.
• deepsearch.py — multi-source deep search aggregation + research briefs + provider settings.
• omnisearch.py — word-or-URL fan-out across every search/scrape/MCP source.
• mcp.py — MCP client (stdio + HTTP) with health monitoring and process-group cleanup.
• local_mcp.py — Local MCP service manager with watchdog and auto-restart.
• diagnosis.py — 30+ read-only health checks.
• manuscript.py — Manuscript assembly + KDP-ready exports (EPUB / DOCX / PDF / HTML).
• server.py — Threaded HTTP server with session auth, audit logging, rate limiting, encrypted keys, optional HTTPS.
• Frontend — hand-written HTML / CSS / vanilla JS (no build step). Includes service worker, SVG cover builder, virtual scrolling, drag-and-drop, inline chapter editor.

1. Start the server: cd app && python3 server.py — it listens on port 8765, seeds the database and admin user on first start.
2. Open http://localhost:8765 in a browser. Log in with the admin credentials (auto-created on first run).
3. Start on Dashboard for the headline numbers, then Niche Library to shortlist categories.
4. Use Validator + Red-Light Scanner before committing to any new idea.
5. Read the Playbook to internalise the rules, then add winning niches to My Picks.
6. Feed your research into Research Vault — upload or fetch; auto-analysis runs on upload. Use Ask the Vault for Q&A with citations, or Deep Web Search + Read & brief for fresh material.
7. In Author Studio, create a book, tick Ground with research, and generate the outline and chapters. Use the inline editor to revise any chapter.
8. Use Omni Search for quick research across all sources, or the AI Toolbox for autonomous research.
9. Re-verify quality from the Testing page — 557 checks run green in an isolated subprocess.

The suite in app/tests/ contains 557 checks across 17 test modules, all green:

• test_engine_db.py — scoring, validation, keyword generation, title ideas, red-light scan, revenue estimation, BSR history, configurable bands, roadmap, undo system, pick dedup, memory PDF, niche CRUD, custom niche evidence/category
• test_api_security.py — API behaviour, auth flow, session cookies, admin user management, ID validation, CSV formula-injection, rate limiting, static containment
• test_ai.py — AI Author Studio in demo mode: outline, chapter, copy-edit, quality, grounding, chat history, streaming, error paths, 429 retry
• test_market.py — Amazon scrape parsing, PA-API preference, caching, throttling, BSR probing, multi-source fallback
• test_market_settings.py — Market data source settings persistence
• test_research.py — PDF/TXT extraction, corpus ranking, web search, URL fetching, auto-analysis, Q&A with citations
• test_deepsearch.py — Multi-source aggregation, provider merging, research brief loop
• test_omnisearch.py — Omni search word/URL fan-out, dedup, AI synthesis
• test_mcp.py — MCP server presets: integrity, tool-pick mapping, stdio/http e2e, preset-mode tests
• test_local_mcp.py — Local MCP service manager: start/stop/watchdog/process cleanup
• test_tools.py — AI toolbox: core tools, dynamic MCP tools, ToolRunner dispatch, function-calling
• test_diagnosis.py — System health checks: server, DB, dataset, AI config, engine probes
• test_manuscript.py — Manuscript exports: EPUB, DOCX, PDF, HTML, MD, TXT
• test_calibration.py — Scoring calibration
• test_master.py — Master update pipeline
• test_service_catalog.py — Service catalog and provider management
• test_live_smoke.py — Live smoke test (expected skip in offline mode)

Run them anytime from the Testing page — the suite executes in an isolated subprocess so your live data is never touched, and the web/AI calls are stubbed so the run never leaves your machine.

• Multi-user authentication — PBKDF2-HMAC-SHA256 password hashing (260k iterations), secure session cookies (HttpOnly, SameSite=Strict, 7-day expiry), admin-only user management. Registration is closed after the first user.
• HTTPS — optional via self-signed certificate: set KDP_HTTPS=1 env var; cert auto-generated on first start. Requires openssl on PATH.
• Encrypted API keys — stored with machine-derived XOR + HMAC (not plaintext in ai_config.json).
• Rate limiting — per-endpoint rate limits, stricter for /api/ai/* endpoints.
• Audit logging — all POST/DELETE actions logged to audit_log table with timestamp, user, action, detail.
• Prompt sanitization — user inputs stripped of control chars, length-capped at 8k, wrapped in delimiters to prevent injection.
• Static file containment — realpath guard prevents path traversal; malformed IDs return 400.
• Local data only — all data in a single SQLite database; nothing leaves the machine unless you configure an outbound service.
• Opt-in outbound calls — Web Search hits DuckDuckGo's keyless API (or your provider), Deep Search hits only sources you tick, Author Studio talks to an LLM only when a key is configured. URLs fetched server-side with scheme allowlist, size cap, no cookies/credentials.
• Service worker — caches static assets for offline access; network-first for API ensures freshness.

Resolved (Phases 1–5, 40 items):

• Token counting, prompt sanitization, chat persistence, demo banner, toast notifications, theme toggle, DB auto-backup, configurable chapter word count, audit log, revenue estimation, BSR history, configurable BSR bands, chapter drag-and-drop reorder
• DB migration system, market cache in SQLite, MCP health monitoring, connection pooling, Playwright retry/fallback, DB busy timeout, undo system (soft-delete + 30s restore)
• SSE streaming chat, inline chapter editor, niche edit UI, virtual scrolling for large lists
• HTTPS with self-signed cert, encrypted API key storage, service worker, SVG cover builder
• Multi-user auth (PBKDF2, session cookies, admin panel, user CRUD, role-based access)

Remaining (future roadmap):

• High priority: Niche time-series tracking (trend charts), PDF Markdown rendering, multi-provider market fallback chain
• Medium priority: Search result caching, top-3 BSR probing, citation management, KENP manual entry
• Low priority: Search history sidebar, ES module split of app.js, bulk operations, keyboard shortcuts, live integration test suite

Full details: INTEGRATION.md in the project root.