Upkeep time changelog

Work a person does by hand every week, handed to the agent instead. This is the Upkeep time slice of the full Docsbook changelog; an entry appears here whenever what it shipped moves this number, whichever part of the panel it landed in.

NEW - 02.09.2026#

Added#

• The MCP server's agent family is now 135 action tools, one per step of documentation work rather than one per discipline. Ten verbs — observe, explain, discover, decide, plan, draft, measure, verify, learn, handoff — across fifteen subjects: your capability map, jobs to be done, topical authority, search intent, programmatic SEO, free tools, original research, AI search, competitors, reader vocabulary, content architecture, internal linking, trust, backlinks and market expansion. Ask for a step (observe_link_graph, decide_next_market, draft_comparison_page) instead of an audit, and get rows you can act on instead of a report. MCP
• Every action tool names the number it is bought to move — support load, organic traffic, AI citations, time to answer, conversion and eight more — in its own description, so an agent choosing between them is choosing an outcome. MCP
• The draft_* tools return the artifact itself — the page, the answer block, the link insertions, the outreach message — as markdown ready to apply, and name the call that applies it. They still write nothing themselves, so the whole family stays safe on a read-only token. MCP
• Generate Issues asks your assistant to look at the project and file what it finds, after two questions: which stage of work you want to be in — observe, understand, discover, decide, plan, execute, measure, verify, learn, coordinate — and which number you want moved. Asking is the point. With no stage named an assistant returns things to build every time, and a backlog with no Measure or Verify in it belongs to a team that never finds out whether the last thing it built worked; the pair you pick also selects the agents that already cover it, so the issues come from this project's own evidence rather than from general advice. Issues

Changed#

• MCP agent pricing is now per tool, not per class. An action tool is priced from the work it declares — how many families of evidence it reads, how many model round trips it may take, whether it leaves your site, whether it writes an artifact — so calls run $0.0740 to $0.2450 instead of a flat $0.2500, and waits run about 20 s to 70 s instead of a blanket "30 s – 4 min". The narrow tools are now cheap enough to call in a loop. MCP
• The 44 previous audit-shaped tools (audit_seo, map_capabilities, diagnose_traffic_drop and the rest) have been replaced rather than renamed. The tools reference lists what took over each one; the four run_docs_* background jobs, audit_geo and the five collect_* collectors are unchanged. MCP
• The landing page now shows the real admin panel above the pricing card, in the same signed-out preview the anonymous draft ships, instead of hand-drawn mockups that drifted from the product they pictured. Landing

NEW - 01.09.2026#

Added#

• A Sources section in the admin panel — what your documentation is allowed to read from. It lists every kind of source Docsbook can read, from your own repository and website to Mintlify, GitBook, ReadMe, Notion, Zendesk, npm and two dozen more, with the ones you have connected lit and the rest offering a Connect. Connect one and asking the assistant to update the documentation or resolve drift starts by fetching it, instead of answering from memory: a repository hands back its files, a site hands back its pages. Sources

NEW - 31.08.2026#

Improved#

• The client picker on MCP tools, Prompts, and a tool's own install card now opens on an Agent tab by default — one prompt to copy for any agent that can read a playbook, instead of hunting your specific client among eight tabs first. The per-client tabs (Claude Code, Cursor, Codex, and the rest) are still there if you'd rather copy the exact command yourself. MCP

NEW - 30.08.2026#

Added#

• Every new project now starts with $1 of real credit, and a few minutes in a card offers $5 more to claim — yours to spend on AI chat, translations, MCP calls or an agent run, with nothing to pay until it runs out. It appears in the sidebar and as a strip across the top of Billing, and claiming it is one button. Billing
• Your agent can now ask one question and get a checked answer back. Nineteen scenario tools each answer a single question about your docs — which pages are one edit away from traffic they already rank for (audit_seo), why traffic fell and what was ruled out (diagnose_traffic_drop), which pages you do not have yet (find_content_gaps), whether a change actually worked (verify_change_impact), whether answer engines can quote you (audit_geo), and fifteen more — each returning a structured answer instead of a paragraph to read. MCP
• Every finding carries the call that would fix it, so an audit hands straight over to run_docs_create, run_docs_manage or run_docs_automate without anyone translating it in between. All nineteen change nothing themselves and work with a read-only token. MCP
• Every tool in the MCP catalog now opens with worked example sentences of its own, where half of them previously showed only a line naming the tool and its arguments. The scenario tools, the background agents, goals and funnels, the assistant's own reports, semantic search and access control all gained three to five phrasings each, plus the chains that hand one tool's finding to the next. MCP
• The public prompt catalog gained two ways to browse them: Audits & diagnosis, for the sentences that ask what is wrong and what the fix would cost, and Background agents, for the ones that start work you come back to. MCP
• write_docs now takes an optional intent, so an agent editing your docs over MCP can record what the person asked for, and get_change_history hands it back along with any prediction attached to that commit. MCP
• An open conversation on the Chat page now has Improve beside Analyze. Analyze reads the transcript for where it went wrong; Improve answers the other question — what to change in your docs so the next reader asking this does not need the chat at all, named at the right layer: a page that should exist, a link that should have connected two pages, or a retrieval problem no rewrite will solve. AI Chat
• Settings ▸ Chat now has two model settings instead of one. AI Visitors Chat Model is what answers your readers; Admin & AI Agent Model is what runs the assistant in your dashboard — the one that reads your analytics, calls tools and edits your docs. Picking a stronger model for yourself and a cheaper one for your readers, or the reverse, is now one choice each. AI Chat
• Ten new scenario tools answer a question about your business rather than about your docs. What every product a buyer considers instead of yours gives away for free, and the need none of them serves (map_competitor_free_offers). Which reader question is answered by a working calculator or validator rather than by a paragraph, and whether it is an existing widget, a custom one, or something needing a service behind it (design_free_tools). Whether a repeating axis in your product justifies a generated page family, and whether a machine can keep that family correct (plan_page_family). MCP
• Forty-six worked example sentences for the new tools in the MCP catalog, including the chains — competitors' free offers into a buildable tool spec into the agent that ships the page, or a support question into the page that closes it. MCP
• Fourteen more scenario tools, one for each method already written in the skills catalog that no tool answered. Why the assistant cannot find an answer that IS on the page (audit_retrieval). Which settings are on and doing nothing, checked against the live site rather than the switch (audit_site_config). Which pages are really tables served as prose, with the widget from your own catalogue that fixes each (design_page_widgets). Which pages the last release made wrong (diagnose_docs_drift). MCP
• And ten more: what should keep happening without anybody remembering, and whether each check belongs in a hook or in CI (plan_automation_workflows); which of these tools your workspace can answer with at all, and the cheapest thing to connect (assess_setup_readiness); the material you already have that could be docs, support answers and community threads included (map_content_sources); whether this kind of change has ever worked here before you repeat it (assess_fix_precedent); which two to four tools your question actually calls for (plan_audit_route); what each number is worth to the business (map_business_value); whether the corpus reads as an authority or as a site that mentions a topic (map_topic_authority); the region readers can only reach from the sidebar (audit_internal_links); which languages are read and which translations are behind their source (audit_translation_coverage); and what shape of answer a query wanted against what the ranking page delivers (diagnose_intent_mismatch). MCP

Changed#

• The billing filters on the MCP tools list now lead with Agent instead of with the cheapest class. The strip scrolls sideways, so at most window widths its tail was off-screen — which put the one family that runs a whole job and hands you back a report where nobody saw it. Sorting the table by billing is unchanged. MCP
• Prompts calling a scenario or background-agent tool now show the PRO badge they always required. Around eighty-five of them were labelled free while the tool behind them was not. MCP
• diagnose_intent_mismatch was being quoted at the wrong price and the wrong wait, because both rate tables matched the bare word "intent" from an older, single-tool rule. It is an agent run and now says so — a caller told to expect a few seconds would have given up on something that takes minutes. MCP
• The MCP tool list was showing the previous build's catalog. Forty-five agent capabilities had shipped, deployed and were answering over MCP while the page listed 101 tools and the Agent filter showed only the four background runners — the list is cached hard because it changes only on deploy, and on a fixed address a stored copy was served without ever being rechecked. The request now carries the build that serves it, so a deploy can no longer be answered with the build before it. MCP
• Assistant in the draft's panel is where you rewrite a page, add one or change the wording, and interactive mode in its message box opens the site with the chat beside it and every block clickable, with a way back to the panel in the corner. AI Chat
• Three feeds joined the built-in roster: Translations (every language generated, outdated or still needed), Language events (which languages readers switch the docs into) and Chat events (questions the AI assistant was asked, where it came up empty, which answers got a thumbs-down). Feeds
• The Tools column in Prompts is now off by default too, alongside Tags. The chips named which endpoints the agent would reach for, and both questions you actually decide on — the price and the payoff — are computed from that same list, with the cost hover naming the dearest thing the prompt touches. Every tool is still listed in the row's card, and the column is one click away in Columns. Prompts

Fixed#

• The public skills catalog now spells names the way the rest of the product does — SEO and GEO rather than "Seo" and "Geo". Skills

NEW - 29.08.2026#

Added#

• Your agent can now hand a whole documentation job to Docsbook instead of doing it itself. run_docs_analyze, run_docs_create, run_docs_manage and run_docs_automate run the matching docs-skill on our side, against your workspace, with the full administrative toolset the skill was written for, and return a run id you read with get_agent_run. Work that takes minutes no longer has to fit in one request, and an assistant with no other Docsbook tools connected can still get an audit done. MCP
• get_agent_run, list_agent_runs and cancel_agent_run report a run's state and live progress, return its report and everything it changed once it finishes, and stop one that is still going. MCP
• Feeds now logs every MCP tool call made against your project alongside the project's own events, showing which tool an agent called, whether it worked, how long it took and what it cost, and filterable by the call's billing class. Dashboard

Removed#

• The Skills section of the admin panel, which listed an externally published catalog on its own release schedule and said nothing about your workspace. The catalog still ships and the assistant still opens skills by name. Dashboard

NEW - 28.08.2026#

Added#

• A Getting started checklist now sits at the bottom of the admin panel's sidebar, showing what your site still needs — its content, your branding, the AI chat, languages, your agent, your domain, and being findable. It ticks steps off as they are configured, collapses to a single row, and disappears once you are done. Dashboard

Changed#

• MCP and Skills are now rows of their own in the admin panel's sidebar, one click each, instead of pages nested inside an Agents section. Dashboard
• MCP and Skills now open on the first tool and the first skill instead of a page about the section, so what your agent can do here is on screen the moment you land. MCP
• Connecting your project is now the first step on every tool and skill page, with the one sentence to paste into your agent and the exact command for your client under it. It used to come after the step it makes possible, and on a skill it could be missing entirely. MCP
• Every tool and skill in the picker now carries an icon, so a list of eighty can be scanned rather than read. MCP
• Text, commands and example prompts on the MCP and Skills pages now scale up on a wide screen instead of staying at phone size. MCP
• Each skill's page still carries the example questions for that skill, next to one command to install it and one line to run it. Skills

Removed#

• The docs-subagents catalog is no longer browsable in the admin panel. Those agents are still installable from the docs-subagents package itself. Dashboard

NEW - 26.08.2026#

Added#

• Cards can now carry a body, their own call-to-action link, and a chosen number of columns or a compact horizontal layout, set on the widget marker. Content Widgets

Fixed#

• Russian-language pages now emit the FAQ and how-to markup that makes them eligible for rich results and AI answers, and a procedure written as a stepper widget is picked up as steps. Widget markers no longer leak into that markup. SEO
• A card whose text contains a link no longer breaks its own layout. Content Widgets

NEW - 25.08.2026#

Fixed#

• The Changes tab no longer stalls on busy repos — a stale per-run cap could stop it recording new commits once a repo shipped docs faster than it drained them. Changes

NEW - 24.08.2026#

Fixed#

• The MCP install commands, raw config and example questions are usable without an account, matching the public MCP page they mirror. Agents
• Opening an MCP tool or a skill no longer hides its catalog for the rest of the session — the breadcrumb goes back, and a failed catalog load retries when you reopen the page. Agents
• Plan names no longer appear to visitors who have no account, on tool descriptions, skill badges or example prompts. Preview

NEW - 23.08.2026#

Added#

• A Widgets gallery in settings shows every content widget your pages can render, each with a preview of the block it produces and a page describing the markdown it expects. Settings
• Any content widget can now be switched off for a project. Its comments stay in your files and every word between them still publishes; only the rich block is withheld. Switch it back on and every page that used it returns, with nothing to re-write. Content Widgets
• Apply to a page on a widget closes settings and turns on editing over your docs, offering that widget first on whichever block you pick. Settings

Changed#

• The live editor and the assistant now offer only the widgets a project has switched on, so neither can write markers that would not render. Content Widgets

NEW - 22.08.2026#

Added#

• get_page_diff_impact returns that same country, language and device breakdown, so an agent can tell a translation-shaped audience from a general rise in traffic. MCP
• The Agents tab has a new MCP page listing every tool this project's MCP server serves — read live from the server, so it is never a stale copy — with each tool's description, its arguments, and the sentences to say to a connected agent to make it fire. Agents
• The MCP page marks the four tools a client can reach with no token at all, so you can see what a reader of your docs could call, not just what you can. Agents
• Needs attention in the Feeds digest counts the events where a reader hit a wall — unanswered chat questions, dead-end searches, stale content and translations, usage limits — separately from routine activity. Feeds
• The Agents tab has a Skills page: every docs skill from the published catalog with its plan gate, install line, the sentences that trigger it, the MCP tools it calls, and its full instructions. Agents

Changed#

• The Agents tab now drills into a catalogue of every docs-subagents agent grouped by pipeline; picking one lists its ready-to-run prompts instead of a single shared MCP-connection card. Agents
• The api widget's playground now takes your workspace's colours instead of a fixed blue: the accent, buttons, focus rings and path parameters follow your brand, and the method chips stay readable on a dark or tinted page. Widgets
• Tool, agent and skill names across the Agents tab read as names (Docs Planner, not docs-planner), with the machine id kept verbatim under each title. Agents

Fixed#

• An api widget endpoint with no example no longer renders its form at half width, and a ### Response block now sits beside the example it should be compared with instead of under the form. Widgets
• Documenting Authorization in an api widget's parameter table no longer renders it a second time as a required query field, which would have put the reader's key in the URL. Widgets

NEW - 15.08.2026#

Fixed#

• Widgets on a generated draft site no longer link to pages that were never created. Draft

NEW - 14.08.2026#

Added#

• Ask the assistant what to improve and the answer is now a list you tick, not prose you re-type. Each row is one concrete change to one of your real pages, or the settings card that applies it; tick several and press Apply once, and they are all done in a single pass. Nothing is ticked for you, and what you leave unticked is never written. The list is drawn from the documentation skill that covers what you asked, what can be measured about your site, and the cards that exist — not from what the model already believed about the topic. AI Chat

NEW - 10.08.2026#

Fixed#

• Switching tabs in the settings widget kept the previous tab's scroll position. Settings

NEW - 07.08.2026#

Added#

• A recommendations widget renders a marked list of findings as cards, ranked by how much each one is costing you. Docs

NEW - 01.08.2026#

Added#

• Editing a page can now add content, not just change it: hovering the seam between two blocks reveals a plus button that inserts a paragraph, heading, list, code block, quote, callout, table or content widget right there. Block Editor
• Your AI agent can now read a public web page and get it back as clean Markdown, so it can check your docs against a competitor's pricing, your own marketing site, or a link that may have gone dead. MCP
• Your AI agent can read and set the call-to-action page through update_branding, and sees it on every workspace it reads. MCP

Changed#

• The admin/reader mode toggle inside an open chat is gone — the mode now follows how you opened the chat (the admin toolbar starts the builder, a site chat widget starts reader mode). AI Chat

NEW - 31.07.2026#

Added#

• A new Stepper content widget renders headed sections as a numbered, connected sequence — for installation guides and multi-stage tutorials where order matters. Content Widgets
• Two new call-to-action content widgets close a page with the next step the reader should take: cta renders a heading and one or two buttons, and cta-form turns the primary action into a single field whose value is carried into the target URL. Both stay compact so a documentation page still reads as documentation. Content Widgets
• A Content Widgets page documents all six widgets, their markdown contract, and how to insert one from the block editor. Documentation
• Generated draft sites now close their selling pages with a call-to-action block, chosen from the widget catalog rather than a fixed list, so a newly shipped widget reaches generated sites automatically. Site Generation

NEW - 29.07.2026#

Fixed#

• Unlock cards now quote the real numbers instead of stale ones: your monthly AI budget in dollars rather than a query count, 15 supported languages rather than "50+", and the actual chat model and MCP tool counts. Billing
• The chat feature is now called AI Chat everywhere in the admin panel, instead of switching between "AI Agent" and "AI Chat" between screens. AI Chat

NEW - 28.07.2026#

Added#

• Your custom questions now appear as clickable suggestion chips when a reader focuses the chat input, and adding a skill swaps them for that skill's own example questions. AI Chat

NEW - 27.07.2026#

Added#

• Filling in a language translates only the missing and outdated pages; pages already up to date are skipped and cost nothing. Translations

NEW - 24.07.2026#

Changed#

• Webhook registration now requires the Business plan consistently, whether registered via the dashboard or an MCP agent. Webhooks
• /llms.txt and the shared preview image now describe Docsbook's current positioning instead of an outdated tagline. SEO

NEW - 17.07.2026#

Added#

• /pricing.md — a plain-markdown pricing page for AI agents to read directly. Pricing

NEW - 14.07.2026#

Added#

• Copy page menu card — independent toggles for each item in the Copy page dropdown (Skills.md URL, view as Markdown, and shortcuts for ChatGPT, Claude, Cursor, Windsurf, VS Code MCP). Content

NEW - 05.07.2026#

Changed#

• Upgrade page no longer shows specific AI-queries-per-month numbers that had drifted out of sync with actual limits. Billing

0.26.5 - 29.06.2026#

Improved#

• Updated landing page feature names for clarity: "AI Agents", "Live Sync", "Auto Translations", "Auto Distribution". Landing

0.26.4 - 12.06.2026#

Improved#

• Buddy mode: Converted /buddy from command to dedicated skill with isolated context — improves modularity and reduces main session token usage.
• Agent daemon: Enhanced reliability with revised auto-commit.sh lock handling and improved logging for task transitions.

0.26.3 - 11.06.2026#

Fixed#

• Usage attribution: When a workspace owner uses the docs-chat widget, their token spend is now correctly charged to the "Admin & AI Agent" category instead of inflating the "Readers (AI Chat)" bar — giving an accurate picture of how much visitors actually cost.

0.26.2 - 11.06.2026#

Improved#

• Agent daemon: Token diet for spawn_session() — now selects model by priority (P1 → Sonnet, P2/P3 → Haiku instead of fixed Sonnet) and adds bash pre-checks in merger role (skip if PR already merged or base=main). Selective directory copy by role (merger copies only routines/ + agents/branch-merger.md instead of full context).

0.26.1 - 11.06.2026#

Fixed#

• Daemon: Unreleased agent:working labels no longer hang forever — added reconcile sweep in sweep_locks() to auto-remove labels without live lock files; also fixed repo context (added -R Docsbook-io/docsbook to all gh-calls) and network hangs (wrapped git/gh in 20/30s timeouts)
• Hooks: auto-commit-hook.sh now removes stale lock files (>10 min) instead of skipping forever after crash

0.26.0 - 11.06.2026#

Added#

• New /chat page with full AI agent for docs: search, edit settings, publish changes, and get answers — all in one conversation interface.
• AdminCard manifest system — all FloatWidget settings tabs now driven by a single ADMIN_CARDS registry, making tabs easier to add and test.
• watch-issues.sh script and local agent daemon for automated pipeline tasks.
• code-scout subagent — investigates code by problem description and creates GitHub Issues with technical context, so Buddy stays in orchestration mode without reading code directly.
• qa-agent now accepts a FOCUS parameter when called directly via Agent tool, enabling targeted testing without a full /qa-plan sweep.

Fixed#

• Agent pipeline agent:working lock now released automatically on any session exit (trap on EXIT in nohup subprocess) — no more manual lock cleanup after agent crashes.

0.25.1 - 08.06.2026#

Fixed#

• Skills: Corrected the npx install link on the /skills page — now points to the correct Docsbook-io/docs-skills package

0.25.0 - 04.06.2026#

Added#

• Onboarding: Interactive 7-step onboarding guide on first login to Docsbook — guided tour highlights key features in FloatWidget toolbar, adapts to user's plan (Free/PRO/PRO+/Enterprise), and remembers when dismissed with hasSeenOnboarding flag in workspaces
• Admin: Fix FloatWidget (toolbar) not appearing for authenticated repo owners after "Start for free" — added direct GitHub repo ownership check in ensureWorkspaceIfMember() so owners see the admin interface immediately
• Skills: SKILL.md schema preview on detail pages (/skills/[name]) — developers now see required/optional frontmatter fields, YAML example from the current skill, and copy-paste instructions before installing with npx docs-skills install

0.24.0 - 04.06.2026#

Fixed#

• Landing: Fix /skills install command showing hardcoded "25 skills" — now uses dynamic index.skills.length (currently 36)

0.23.0 - 03.06.2026#

Added#

• Growth: New /enrich-audience command for the first growth-reasoning team — reasons over about/ + Axiom analytics and appends insights back into the product business-layer. Adds cross-artifact drift contract in CLAUDE.md + AGENTS.md (MCP ↔ docs-skills ↔ docs-subagents ↔ docs-claude-plugins dependency graph)

0.22.2 - 28.05.2026#

Changed#

• Official documentation now served at docsbook.io/docs — middleware rewrites /docs/* internally instead of redirecting to docsbook-io.docsbook.io; canonical URLs, sitemap, JSON-LD, and all links updated across landing, admin, and MCP pages

0.22.0 - 28.05.2026#

Removed#

• Removed server-side Source of Truth indexing — get_doc_graph, read_doc_sections, reindex_doc_graph and the 17 doc_* LSP-style MCP tools are gone. Graph search now runs locally via the docs-claude-plugins package (/plugin install docs-sync@docs-claude-plugins). Deleted src/lib/source-of-truth.ts, src/lib/mcp/lsp-tools.ts, the reindex REST route, the daily stale-check cron and the two smoke scripts

0.21.3 - 25.05.2026#

Fixed#

• Mobile /skills and /mcp pages: added hamburger mobile menu to Header with full nav links, "Start for free" and "Log in" CTAs
• Mobile /skills top padding: reduced from pt-28 to pt-20 on mobile (consistent with /mcp)
• SkillsInstallSelector: install/use columns no longer stack on tablets — now lg:grid-cols-2
• SkillInstallGuide: install/use columns now split at sm: breakpoint for earlier two-column layout

0.21.1 - 25.05.2026#

Added#

• Short marketing alias docs.docsbook.io for the product documentation — opens the same content as docsbook-io.docsbook.io/docs/* without redirect (URL stays clean in the browser). New DOCS_ALIAS_SUBDOMAINS map in src/proxy.ts rewrites docs.docsbook.io/{path} → /docsbook-io/docs/{path}; /api/* is passed through untouched, original subdomain URLs keep working.

0.20.2 - 25.05.2026#

Added#

• UTM parameters on all internal CTAs leading from /skills, /mcp, /docs (Preview banner), and the blog to the landing page — every /start and /connect link now carries utm_source (skills / mcp / preview / blog), utm_medium (nav / cta / banner), and utm_campaign (e.g. header_signup, mcp_start_free_top, preview_connect, post slug for blog). New src/utils/utm.ts helper (withUtm()) wires the landing Header/Footer via an optional utmSource prop, the two inline CTAs on /mcp, and the PreviewConnectBanner on workspace pages. Blog post CTAs (docusaurus_vs_docsbook, mintlify_vs_docsbook, gitbook_vs_docsbook, ai_search_documentation, documentation_seo_guide, how_to_host_docs_from_github, why_documentation_matters) now tag their conversions per post. The landing page itself stays UTM-free so internal scroll-to-CTAs aren't mis-attributed

0.20.1 - 25.05.2026#

Changed#

• Landing page positioning rewritten for AI crawlers — ChatGPT and Perplexity were describing Docsbook as a plain GitBook/Mintlify/Docusaurus alternative, missing the entire AI-Native layer. Hero H1 changed from "The AI Knowledge Platform" to "Docs from GitHub. For humans and AI agents." with concrete subtitle naming MCP, llms.txt, and 15 languages. New full-width "Built for AI agents" bento card with terminal mock (claude mcp add), MCP tool grid (doc_outline, doc_search_text, read_doc_sections, …) and client logos (Claude Code, Cursor, ChatGPT, Perplexity, Cline). New "AI Agents" social-proof tab with CTA to /mcp. metadata.title, metadata.description, JSON-LD SoftwareApplication.featureList, and FAQPage rewritten to surface MCP server, llms.txt, Source of Truth graph, Skills catalog, and updated pricing ($150 lifetime PRO / $59/mo PRO+) so AI search engines cite the current product correctly

0.20.0 - 25.05.2026#

Added#

• SEO content hub — 20 new long-tail GEO/AEO blog posts in docs/blog/ targeting AI search citation (ChatGPT, Perplexity, Claude, Gemini) and high-intent developer queries. Covers comparisons (Docusaurus vs Docsbook 2026, AI docs platform comparison, free hosting comparison, docs as code vs managed), AI infrastructure (llms.txt complete guide, JSON-LD for documentation, MCP server for documentation, docs-skills for AI agents, how to get docs cited by ChatGPT, Perplexity citations for docs, multi-language documentation SEO, AI chat build vs buy), migrations (GitBook → Docsbook, Docusaurus → Docsbook), and practical guides (custom domain how-to, API documentation best practices 2026, documentation analytics, README → docs site, why README-only projects need a docs site, best docs platforms for startups 2026). docs/blog/README.md restructured into five sections: Foundations, SEO & AI search (GEO/AEO), AI features, Comparisons & migration, Practical guides

0.19.0 - 25.05.2026#

Added#

• MCP visitor activity drill-down — two new tools on PRO+ (get_top_visitors and get_visitor_activity) let AI agents investigate what one specific anonymous visitor actually did end-to-end. get_top_visitors returns the most active anonymous visitors with a stable hashed visitor_id, pageview count, country, and first/last seen; pass that visitor_id to get_visitor_activity to get the full chronological event timeline (pageviews, page feedback, CTA clicks) with paths and event-specific details (vote, query, href, heading, …). get_page_journeys also returns the same visitor_id so journeys can be drilled into immediately. visitor_id is sha256(VISITOR_ID_SALT + repoFullName + ip).slice(0,16) — stable across sessions for the same person within one workspace, but raw IPs never leave Axiom

0.18.0 - 24.05.2026#

Added#

• Devices, Browsers and AI Visits analytics — new row of cards under Pages/Referrers in the Analytics tab. First card has tabs for Devices (Mobile/Desktop/Tablet) and Browsers (Chrome, Safari, Firefox, Edge, Brave, Arc, Vivaldi, Yandex…) with favicon icons. Second card lists AI crawler visits (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bingbot, Applebot-Extended, Meta-ExternalAgent, CCBot, Bytespider, MistralAI-User and 12+ more) grouped by provider so you can see exactly which AI agents read your docs
• Sitelinks-friendly structured data on the landing — added SiteNavigationElement JSON-LD for 8 key sections (Quick Start, AI Features, MCP Server, Agent Skills, Documentation, FAQ, Blog, Changelog), an ItemList with top destinations, and WebSite.hasPart linking the main pages so Google has explicit signals for generating sitelinks under the docsbook.io result
• New sitemap entries — /mcp and /skills with priority 0.9, plus /connect with 0.5, so Google can discover and weigh these promo pages
• Simplified install/use guide on each skill page /skills/[name] — tabs for 7 AI clients (Claude Code, Cursor, Codex CLI, Windsurf, Cline, Gemini CLI, Copilot), two steps (Install + Use) with the command pre-filled for this specific skill, plus a runtime-discovery block via Docsbook MCP
• New opinion blog post /blog/notion-for-docs-engineering-lessons — first-person engineering essay on why Notion stops working as a docs system once docs leave the building (SEO surface vs internal wiki, version control drift, multilingual coupling, AI crawler discoverability, performance budget, export lock-in, wiki-vs-docs permission split) with a soft Docsbook pitch in the closing section; written for SEO ("notion for documentation") + outreach + objection handling
• Twitter teaser thread for Product Hunt launch at marketing/twitter/ph-teaser-thread.md — 9-tweet building-in-public thread (D-10 hook + 7 building-in-public tweets covering Anonymous MCP, llms.txt auto-discovery, TOON format, Docusaurus alternatives guide, attribution tracking, sitelinks JSON-LD, skills install UX + CTA), each tweet ≤280 chars, character counts inline, posting notes with UTM campaign ph-teaser-twitter

Changed#

• Moved "Get Support" out of the admin sidebar — replaced the bulky "Help & Support" section with a subtle "Need help? Contact support" footer link pinned to the bottom of the settings modal sidebar, freeing vertical space

Fixed#

• AI Skills cards in the admin no longer 404 on workspace subdomains — clicking a card now opens an in-place modal with the full SKILL.md (description, install snippets for 7 AI clients, keywords, MCP tools, GitHub link) instead of routing to /skills/<name> which only exists on docsbook.io. Landing-page behavior is unchanged

0.17.3 - 23.05.2026#

Changed#

• Reworked landing header navigation — replaced old category dropdowns (AI, Analytics, Branding, Widgets, Translation) with 3 direct links (AI, MCP, Skills) plus 2 curated dropdowns: Documentation (Quick Start, Basics, Creating Docs, Custom Domain, AI Translations, FAQ) and Blog (all 5 posts)

0.17.1 - 23.05.2026#

Fixed#

• Prevent race conditions in monthly usage limits for AI Chat, Translations, and Reindex — concurrent requests could each pass a stale pre-check and push counters past the plan limit (visible as 78/50 pages translated on Pro). Replaced check-then-act with atomic conditional UPDATE ... RETURNING in batchTranslate, /api/ai-chat, and the MCP reindex endpoint

0.16.0 - 22.05.2026#

Fixed#

• Auth redirect loop after GitHub OAuth — stale callbackUrl cookie pointing at a subdomain /connect caused an infinite redirect cycle; NextAuth redirect callback now normalises any subdomain /connect → docsbook.io/connect

Added#

• Subscription management UI in FloatWidget pricing tab — shows current plan, subscription status, next billing date, and Manage subscription button linking to Paddle Customer Portal (Pro+ only)

0.15.0 - 22.05.2026#

Improved#

• Search widget UX with breadcrumb paths and "Ask AI assistant" option in Search Bar

0.13.0 - 18.05.2026#

Added#

• Documentation graph indexing for AI agents in Source of Truth (Enterprise)
• Source of Truth settings panel in FloatWidget (Enterprise)

0.11.0 - 16.05.2026#

Fixed#

• Default theme now always applies when theme-switching widgets are disabled

0.10.0 - 15.05.2026#

Added#

• Bring-your-own API key for OpenAI, Gemini, Anthropic, OpenRouter in AI Agent (Pro/Enterprise)

Fixed#

• AI Agent settings always shows 3 question input fields

0.2.0 - 08.05.2026#

Fixed#

• AI Agent tab now loads with a single request instead of three