
2 · configure
3 · deploy
ProAvailable product — files, demo URL and secure download get mapped in EDF 0.3.
Ask-the-docs chat with a dashboard for what your docs are missing
Everything in DocsChat Widget, plus what you need once real visitors use it. An owner dashboard shows what people ask most, the questions your docs could not answer (with CSV export), thumbs-down comments, response times and cache hits. Optional semantic search with embeddings (OpenAI-compatible, including local Ollama) blends meaning with keywords and keeps off-topic questions out. A "talk to a human" form sends the question and the conversation to your Slack, Discord, Teams or signed webhook. A crawler turns a website or sitemap into Markdown. Several doc sets on one server, an injection guard, personal-data masking before anything is stored or sent to the model, an answer cache, and a check of every AI answer for invented or missing citations. Widget in English, Polish, German, Spanish, French and Italian. TypeScript, zero runtime dependencies; Vercel functions or one Node file.
Origin: EWDS Store (epicwebdesignstudio.com) · Roadmap window: EDF 0.2


Placed in the private vault layout (deploy/vault/products). Downloads stay disabled until checkout is connected.
Derived by the mapping engine (family · layer · tags · dependency rules). Stored in the catalog as compatibility, dependencies, relations[].
| VERSION | STATE | NOTES |
|---|---|---|
| v2.0.0 2026-10-03 | AVAILABLE | Current release. This package contains: Everything in DocsChat Widget: Markdown to search index in one command, ~15 KB Shadow-DOM widget with streaming, citations, suggestions, copy, retry, Stop and dark/light themes, serverless endpoint plus standalone Node server, Claude or any OpenAI-compatible model or no AI at all, origin allow-list, rate limits and an off-topic gate that skips the model; Owner dashboard (/admin.html behind a token): most-asked questions, gaps in your docs with CSV export, rated-down answers with comments, weakly grounded answers, response time, cache hits, most-cited pages; questions stored redacted, no IP or cookie, 90-day retention; Semantic search: embeddings from any OpenAI-compatible API (OpenAI, Ollama, LM Studio), int8-packed vectors, incremental rebuilds that only embed edited passages, hybrid ranking and a similarity gate; `npm run eval` measures it and suggests the threshold; Talk to a human: form in the widget posts the question and recent conversation to your webhook (Slack, Discord and Teams work as-is, optional HMAC signature), honeypot, per-visitor and daily limits, retries; Website crawler: URL or sitemap to Markdown, obeys robots.txt, one host, polite delay, conditional requests, only changed pages rewritten, optional prune; Collections: several doc sets on one server, chosen with data-collection. |
| → EDF 0.2 | Roadmap | Target window for the next stage. |
Verified with 102 automated tests, a clean-room install from the zip (npm ci, tests, build, server answers), and a real end-to-end run on a Node server with a local Ollama model (llama3.2:3b) for answers and nomic-embed-text for embeddings: cited and verified answers, cache hit, semantic path, off-topic and injection refusals, redacted analytics, signed webhook to a local receiver, admin API, CORS, plus a headless-browser pass over the demo. On the 31 sample questions, embeddings raised top-1 from 84% to 92% and off-topic rejection from 4/6 to 6/6 (top-3 went from 100% to 96%); the sample synonyms were tuned on that set, so expect lower numbers on your docs. The Anthropic, OpenAI and Gemini-style calls, the Vercel deployment, Upstash, and delivery to a real Slack/Discord/Teams webhook are tested against fakes, not with real accounts. The answer check is lexical: it catches invented citation numbers, missing citations and unsupported sentences, not every wrong claim. The crawler reads static HTML only (no JavaScript rendering) and was tested on a local test site. The live demo uses an offline stand-in for the model and an on-page inbox, and says so. Requires Node 20.19+.
Standard (personal or commercial use in one end project or your own business) · Extended ×3 (SaaS, unlimited client projects, finished products only). Agency licence comes later.
Personal: use on websites and docs you own. Commercial: use for one client or company. Extended: unlimited sites, clients and products. No resale, redistribution or hosting as a service for third parties.
Activated by 1.0 (download) and 2.0 (one-click deploy). Download will be a 10-minute signed URL from the R2 vault — never a public /files/product.zip.



