Version history
Every update, from day one.
Track every release — from chatbot and MCP improvements to scheduled automations, trusted plans, and safer WordPress operations.
Claude Sonnet 5.5, on launch day.
PressBot 1.12.0 adds Anthropic’s new Sonnet the day it shipped, at the same price as Sonnet 5, and makes it the Admin Agent default.
Claude Sonnet 5.5
claude-sonnet-5-5at $2 / $10 per million tokens for visitor chat, the Pro Admin Agent, Shield, and the blog pipeline- New default for the Admin Agent and the pipeline’s Writer and Critic. Models you already chose stay as they are; Sonnet 5 remains selectable as previous
- The visitor chatbot default stays Claude Haiku 4.5
Under the hood
- Sonnet 5.5 no longer accepts the setting PressBot used to turn thinking off. PressBot now sends the lowest setting it does accept, so visitor chat and short pipeline steps stay fast and within budget
Every line readable, every outside call listed.
PressBot 1.11.2 ships the readable source for its JavaScript and documents every outside service it can contact.
- The plugin includes the TypeScript source and build config behind its JavaScript, so you can review or rebuild it
- The readme lists each outside service (your AI provider, optional video and knowledge endpoint, opt-in error reports, Freemius), what it receives, and when
- Tested up to WordPress 7.1
A Yoast audit that ranks what to fix.
PressBot 1.11.1 makes the Pro Yoast SEO audit catch what Yoast flags one post at a time, across the whole site, worst first. It also fixes the MCP setup instructions in Settings.
Yoast SEO audit
- Flags meta descriptions of 120 characters or fewer, or over 156, the same range Yoast uses
- Finds titles and descriptions duplicated across posts, and says which posts share them
- Filters by status and category, works through large sites in batches, and lists the posts that need the most work first
MCP setup
- Settings now shows working, click-to-copy snippets for Claude Code, Cursor, and Codex, filled in with your site URL and username. The old instructions pointed Claude Code to a file it doesn’t read
Your chatbot answers from your pages. Free.
PressBot 1.11.0 lets the free visitor chatbot search your published pages and posts, answer from what they say, and link the page it used. Nothing to index and no extra service: it works with all five AI providers on your own key.
Site answers
- Searches your published pages and posts, and reads a whole page when a list or set of steps is cut off
- Links the page it used, and source cards show only the pages the answer relies on
- Choose which content types it searches, in priority order. Password-protected, noindex, and excluded pages are never used
- On for new installs. Existing sites turn it on with one click from the notice in the PressBot admin
Smoother answers
- If the chatbot can’t finish an answer after using its tools, visitors get links to the pages it found instead of an empty reply
- With focus mode on, anything your site has published about counts as on-topic
Gemini visitor chat finishes its answers.
PressBot 1.10.3 fixes Google Gemini visitor chat when a question needs a tool, like a product search or a booking check, and keeps the visitor chatbot off the web.
Fixed
- With Gemini 3 models, visitor questions that used a tool (WooCommerce products, booking availability, External Knowledge) now get an answer instead of an error
- Gemini replies no longer come back blank when a message arrives in pieces
- With Claude, the Admin Agent’s web search setting no longer reaches the visitor chatbot, so visitors can’t trigger paid web searches or get answers from outside your site
Gemini chat works on WooCommerce stores.
PressBot 1.10.2 fixes visitor chat on WooCommerce sites that use Google Gemini.
Fixed
- Gemini rejected every visitor chat on WooCommerce sites because of how PressBot sent the product-categories tool. It now answers normally. The Admin Agent was not affected
Gemini errors that say what's wrong.
PressBot 1.10.1 fixes a misleading message for Google Gemini users. A rejected API key used to ask you to rephrase your request.
Fixed
- A rejected or expired Gemini key now says so and explains how to replace it, and sends the invalid-key alert
- If Google doesn't offer the Gemini API in your server's region, PressBot now tells you
- For other Gemini errors, administrators also see Google's own error text. Visitors still see a plain message
CTR Doctor: fix the snippet, then check the result.
PressBot 1.10.0 lets CTR Doctor fix what it finds. Ask the Pro Admin Agent to rewrite your search titles and descriptions, approve the plan, and PressBot checks back about a month later to see whether more people clicked.
Fix plans you approve
- New titles and descriptions for up to 10 pages go into a Trusted Plan. Nothing changes until you review, edit if you like, and apply it, and every change has a best-effort restore
- Writes through Yoast SEO when it's active. On sites without an SEO plugin, the new PressBot SEO fields give each page its own search title and description
Results, measured
- About 34 days after you apply, PressBot compares Search Console click-through on the same searches at similar positions, so a ranking change isn't mistaken for a better snippet
- The plan then says improved, declined, no clear change, or not enough data. A result counts only if it's statistically significant
CTR Doctor: find the clicks you're missing.
PressBot 1.9.4 adds CTR Doctor to the Pro Admin Agent and MCP. Ask which titles to rewrite, and it ranks pages by the clicks they're likely missing for searches they already rank for, next to the title and description Google shows today.
What it finds
- Each search is scored at its own ranking, so a strong page-one query isn't hidden by a page's weaker average position
- Flags titles that may be cut off, missing descriptions, and titles that don't mention the page's top search. Leave out brand searches to see the real picture
- Pages just past page one are labelled separately: there, better rankings matter more than better wording
Honest numbers
- Read-only. Missed clicks are estimates from a CTR-by-position benchmark, not a forecast
- The Search Console report now says when position or impression limits narrow its rows but not its totals
Ask Search Console better questions.
PressBot 1.9.3 lets the Admin Agent and MCP clients filter and sort Search Console data, so you can ask which searches reach a page, or which page-one results get few clicks. Gemini setup now says what Google's Free tier is for.
Search Console report
- Filter by page (a full URL or a path like
/pricing/), by query, or exclude a query such as your brand name - Sort by impressions, CTR, or position, and keep only rows in a ranking range or above an impression threshold. Reports say when a long list may be incomplete
Clearer Gemini setup
- Google's Free tier is described as a way to test. For a live or client site, PressBot recommends a paid Google project: higher limits, and Google doesn't use your chats to improve its products
Ollama Cloud, with your own key.
PressBot 1.9.2 adds Ollama Cloud in early access. Connect your own key to Ollama's hosted API, then choose a model for the visitor chatbot or Pro Admin Agent. We tested a live Gemma 4 public-chat conversation; Admin Agent tool use and other models are open for feedback. Model quality, availability, and usage costs vary by Ollama plan.
Hosted Ollama choices
- Five curated visitor-chat models, starting with Gemma 4 31B; nine Admin Agent choices, including GPT-OSS 120B and Kimi K3
- Public streaming and tool-call continuation; separately encrypted Ollama API key; connection status updates as soon as Settings saves
Choose by role
- Getting Started now advises testing models for the public chatbot, Admin Agent, and content pipeline, and reviewing outputs before publishing or changing a site
- Hosted API only: no local Ollama server or custom endpoint. Ollama is not an option for Shield or the pipeline's web-search strategist; image generation stays on Gemini or OpenAI
GPT-6 Luna. GPT-6 Sol. No forced migrate.
PressBot 1.9.1 is a catalog-only patch on 1.9.0. GPT-6 Luna (gpt-6-luna) is the fast, affordable OpenAI pick at $0.10/$0.50 per MTok. GPT-6 Sol (gpt-6-sol) is the balanced agentic pick at $2/$10 per MTok for both visitor chat and the Admin Agent — unlike GPT-5.6 Sol, it is not agent-only. GPT-6 Astra stays the OpenAI flagship. GPT-5.6 Luna, Terra, and Sol remain selectable as previous. Saved settings are not auto-migrated. Defaults stay Haiku public / Sonnet agent.
GPT-6 Luna
- GPT-6 Luna (
gpt-6-luna) at $0.10 input / $0.50 output per MTok — fast and affordable - Available for visitor chat, the Admin Agent, Shield, and session assist
GPT-6 Sol
- GPT-6 Sol (
gpt-6-sol) at $2 input / $10 output per MTok — balanced agentic - Available for visitor chat and the Admin Agent (
agent_onlyfalse), unlike previous GPT-5.6 Sol
Previous models stay put
- GPT-6 Astra remains the OpenAI flagship. GPT-5.6 Luna, Terra, and Sol stay selectable as previous; labels now say Previous
- Saved settings are not auto-migrated to GPT-6. Plugin defaults are unchanged (Haiku public / Sonnet agent)
- Session translate/summarize prefers
gpt-6-lunaon the cheap OpenAI ladder; Terra remains the higher-quality OpenAI option
Modern layouts. Home. A video that waits for Play.
PressBot 1.9.0 adds an optional Modern presentation for the visitor chatbot and the Pro admin agent — independently. Compact 380×520, Comfortable 440×680, Large 520×760, constrained to the viewport; mobile stays responsive. Optional Home with Home/Messages nav. Featured YouTube is public-only. Fresh installs start Modern; existing sites keep Classic until an admin opts in.
Modern layouts & Home
- Independent Classic/Modern for public chat and the Pro admin agent. Modern sizes: Compact, Comfortable, Large; viewport-constrained; mobile is responsive, not a fixed desktop size
- Optional Home screen plus Home/Messages nav. Home has welcome copy, a start/continue action, and existing conversation starters. Admin Home also lists recent conversations
- Basic settings stay layout, size, and Home on/off. Collapsed Customize: intro label, multiline title, title-size presets Small/Balanced/Large, description, conversation-button text. Reset home content preserves layout, size, and Home. Conversation greeting stays separate from the Home title
- Modern opens with a short scale/slide from the launcher; reduced-motion disables it. “Answering about” is now an upright 16px document SVG
Public-only featured video
- Optional YouTube card on the public widget only: title, description, custom thumbnail from the Media Library or an HTTP(S) image URL. No mascot or brand poster. Blank or broken thumbnail falls back to a simple title/play card
- Player uses youtube-nocookie.com and loads only after Play. Leaving Home or closing the widget removes the iframe. Thumbnails may load before Play. No video on the admin agent
Claude Opus 5.5
- Claude Opus 5.5 (
claude-opus-5-5) is available for the Pro admin agent and Shield at $4/$20 per MTok. Opus 5 remains selectable as previous. Visitor chatbot defaults are unchanged
Existing sites stay Classic
- Fresh installs seed Modern / Comfortable / Home on for both surfaces. Existing sites keep saved choices; missing layout settings resolve to Classic. Reactivation does not opt an existing site into Modern. No domain-specific override for pressbot.io
- Dismissible invitation on PressBot pages for Classic visitor layouts. Switching back to Classic remains available
Gemini & setup (bundled 1.8.9)
- Gemini 3.8 Flash is the default for new Google-provider selections; Gemini 3.5 Flash Lite remains available. Saved model choices are not globally migrated
- Free-tier eligibility depends on model, project/quota, and request region. In-product copy no longer claims unconditional “no credit card” or a fixed evergreen free quota
- Getting Started and Conversations headings no longer show unconditional Pro badges in the free edition
Budget agent. Drop knowledge in. Translate any session.
PressBot 1.8.8 unlocks DeepSeek-V4.1-Flash as a Budget Admin Agent option (and content-pipeline steps) — Sonnet/Haiku stay the defaults; Strategist remains Anthropic/Gemini-only for web search. Settings Knowledge Base accepts drag-and-drop .md / .txt notes. Conversations get free per-session translate and summarize with a model picker from whatever keys you already configured.
DeepSeek Budget Agent
- DeepSeek-V4.1-Flash (
deepseek-flash) is selectable for the Admin Agent and content-pipeline steps — labeled Budget - Strong fit for Researcher / Categorizer; optional for Writer / Critic. Defaults unchanged
- Strategist stays Anthropic/Gemini-only (web search). Provider copy now covers public chat, Shield, and Budget agent/pipeline
Knowledge dropzone
- Settings Knowledge Base accepts drag-and-drop or click-to-browse for
.md/.txt(multi-file, 500 KB each) - Imports create notes via the existing composer API; typed composer unchanged. Note cards refresh immediately after import
Session translate & summarize
- On Conversations → Session Detail, translate the full transcript or generate a short admin summary into a chosen language (free)
- Model picker uses configured providers (DeepSeek Flash, Luna/Terra, Gemini Flash Lite/3.8, Haiku/Sonnet), cheapest first. Choice is remembered
Quiet Shield has a home. Knowledge reads like notes.
PressBot 1.8.7 puts Quiet Shield on both surfaces from one shared card. On Shields, Shield Settings comes first and Quiet Shield follows in the same save form. Settings keeps a compact twin with a deep link so the UIs cannot drift.
Settings → Knowledge Base browses notes as readable cards and adds or updates them through a message-like composer. Markdown file upload stays collapsed for power users. The September catalog adds GPT Image 2.5, Gemini 3.8 Flash, and DeepSeek-V4.1-Flash.
Quiet Shield
- Shared Quiet Shield card: Shields is the home panel; Settings keeps a compact twin with a deep link to the full controls
- On Shields, Shield Settings comes first and Quiet Shield follows in the same save form
- Urgent vs digest, today’s interrupts vs daily budget, and mute status live on the card — alert-type checkboxes stay under Settings → Telegram
Knowledge Base
- Browse notes as cards: human title, snippet, Edit / Delete. No filename, size, or
.mdin the primary UI - Add or update from a message-like composer (optional title + body). Markdown file upload stays collapsed for power users. Storage is still
.mdunder the hood
September catalog
- GPT Image 2.5 Flare (
gpt-image-2.5-flare, fast default) and Sunburst (gpt-image-2.5-sunburst, higher precision), with GPT Image 2 kept as the previous-generation option - Gemini 3.8 Flash (
gemini-3.8-flash) is the Gemini default for agent/vision fallbacks and appears in public, agent, and Shield selectors - DeepSeek-V4.1-Flash (
deepseek-flash) as the canonical DeepSeek model for public chat and Shield. Native multimodal (vision) - Anthropic, Gemini, OpenAI, and DeepSeek IDs/labels refreshed. Claude Fable 5.1 and GPT-6 Astra added as agent-only flagships; Gemini 3.7 Flash is labeled as the previous agentic Flash. Legacy DeepSeek Flash IDs remain as hidden compatibility aliases
Admin clarity
- Shields page maps Shield vs Quiet Shield vs Sentinel; Settings Shield/Telegram copy now points to the primary surface instead of implying every security control lives in Settings
Agents can connect without an LLM key
PressBot 1.8.6 ships an agent-recommendable surface: provider-free, read-only MCP maintenance tools so an agent can inspect a WordPress site without bringing an LLM key. Discovery docs ship in the plugin zip. The free MCP catalog is now 13 tools (from 8), the full set is 115 (from 107), and free-tier MCP enablement actually sticks.
Agent surface
- Provider-free read-only MCP maintenance tools: baseline, recent changes, change timeline, health probe, and change-risk assessment
- Agent discovery docs in the plugin zip:
docs/for-agents.md,docs/for-agents.json, and a WordPress maintenance skill
Free MCP catalog
- Tool count increased to 115 (from 107); MCP free tools increased to 13 (from 8)
- Free MCP maintenance tools no longer depend on premium-only ToolExecutor, so the 13-tool free catalog works in the free build
Fixes
- Free-tier MCP enablement now persists correctly
Quiet Shield, a local black box, and category tools that wait for you
PressBot 1.8.5 (Pro) makes Telegram security noise earn the interruption. Quiet Shield keeps routine blocked-login chatter in the daily digest, while real-account attacks and high/critical Guardian findings still fire with a clear “why this needs you” reason. Sentinel Local Black Box records observed changes on the site itself. Category correction tools for MCP and the Admin Agent require confirmation with preview, snapshot, and restore.
Shield & Sentinel Pro
- Quiet Shield — severity-based Telegram delivery. Routine blocked-login noise stays in the daily digest; real-account attacks and high/critical Guardian findings still interrupt. Deferred urgents use a structured digest queue (3-day retention, urgent preferred)
- Temporary Mute 1h from Telegram security alerts, plus a configurable per-site daily immediate-alert budget (default 10; Shield Settings shows usage). Budget resets at the site timezone midnight
- Sentinel Local Black Box — opt-in from PressBot → Shield → Sentinel. Records observed changes (plugins, themes, options, posts), runs bounded health probes (~20s typical), and stores derived hypotheses. Evidence Ribbon shows causal traces. 7/14/30/90-day retention. Local-only, no HQ yet
Category tools & models
- Category correction for MCP and the Admin Agent:
update_postsupportscategories_mode=append|replace, plusremove_post_categoriesandswap_post_category. All require confirmation with preview, snapshot, and restore - DeepSeek V4 Flash support for the public chatbot and Shield security analysis (ultra-low-cost option)
- OpenAI support for Shield security analysis (previously admin-only)
Improvements
- Plugin Guardian escalates severity based on patterns introduced by the update, not merely present. Badge and verdict sentence now agree. Product-behavior focus over fear copy
- Model surfaces updated — DeepSeek V4 Pro removed from user-facing agent/security selection. OpenAI and DeepSeek Flash added to the Shield selector
- Bogus distributed scan campaigns aggregate into the digest instead of Telegram; urgent campaigns remain retriable after mute/budget clears
- Tool count increased to 107 (from 105); MCP free tools increased to 8 (from 7)
Fixes
- Daily digest restores the Quiet Shield queue unless every authorized Telegram recipient accepts the message
- Plugin Guardian high findings no longer use “critical” action-reason wording
- Guarded
PRESSBOT_TOOL_COUNT/PRESSBOT_MCP_FREE_TOOL_COUNTdefines to avoid collisions with marketing themes that define the same constants
Sources you can trust, feedback you can act on
Public answers can now show compact source cards built only from validated retrieval results and same-site links the server resolves to published content. The browser never trusts model-emitted URLs on its own.
Thumbs-down opens a short recovery prompt with optional visitor notes stored separately from AI-visible messages — surfaced in conversation detail, CSV export, and backups.
Page-aware chat, a draft pipeline for MCP, and Search Console
The public chatbot now keeps the conversation with you as you move between pages on the same site, with a visible Answering about bar for the current page and a clear Start over control. Each turn stores a canonical source URL (no query strings or fragments) so operators can see where a visitor asked from.
On Pro, run_content_pipeline is available to MCP clients and the admin agent. Pass a writing prompt; PressBot queues research, write, critic, optional image, and categorize, then always saves a draft — never publishes. On MCP, the call returns a job id immediately so long runs do not hit client timeouts; poll status until the draft is ready.
Also on Pro: Google Search Console reporting reuses your existing GA4 service account. Query top searches and landing pages from the admin agent or MCP via get_search_console_report, and fold organic-search evidence into composite site reports.
Stability fix
Improved how the admin and widget scripts are cached, so the settings screen loads reliably right after an update.
Claude 5, GPT-5.6, and DeepSeek V4 Pro
PressBot 1.8.1 refreshes every provider’s model list. Anthropic gains Claude Sonnet 5 and Claude Opus 5, OpenAI moves to the GPT-5.6 family (Luna, Terra, and Sol), and DeepSeek adds V4 Pro. If your site is pinned to a model that has since been retired, PressBot moves you to the closest current model from the same provider automatically — nothing to do on your end.
Models
- Anthropic: Claude Sonnet 5 (balanced) and Claude Opus 5 (most capable); Claude Haiku 4.5 stays the fast, low-cost default for the visitor chatbot
- OpenAI: GPT-5.6 Luna, Terra, and Sol. Luna is now the cheapest option on the list at a fraction of a cent per conversation
- DeepSeek: V4 Pro joins V4 Flash for agent work
- Google: the Gemini 3 family carries over unchanged
Reasoning where it helps
- The admin agent now lets Claude 5 models reason before acting, which improves tool selection on multi-step tasks
- The visitor chatbot deliberately does not, so replies stay fast and per-conversation cost stays predictable
- Fixed an internal version constant that still reported 1.7.1
Find out whether AI actually cites your pages
Auditing your content for AI readiness is guesswork until you ask a real model. The AEO Live Simulator generates realistic questions for a page, asks your own AI model with live web search, and reports whether your page gets cited — with the evidence for each answer. Gaps become a reviewable Trusted Plan in one click, and you can re-run afterwards to confirm the fix worked.
AEO Live Simulator Pro
- Generates 3–5 realistic user questions per page and asks them with live web search
- Reports cited, partially cited, or not cited, with the supporting evidence per question
- Turns the gaps into a Trusted Plan you review before anything changes
- Re-run to compare against the previous result
Monthly AI budget cap
- Set a monthly budget for your estimated API spend
- When the estimate crosses it, the visitor chatbot can switch to your provider’s cheapest model automatically
- The admin agent is never silently downgraded — you get a dashboard notice instead
- Analytics shows a budget progress bar and what Anthropic prompt caching saved you this month
It learns your site — but only what you approve
After a Trusted Plan runs, PressBot can suggest reusable guidance it picked up from the outcome. Every suggestion waits in Automations → Learning until you approve it, so memory stays something you can inspect and control rather than quiet self-training. This release also cuts repeat API costs through Anthropic prompt caching.
Operator-controlled memory
- New Automations → Learning screen for reviewing, approving, or dismissing memory suggestions
- Suggestions come from applied or cancelled Trusted Plan outcomes; nothing is saved without your approval
Compatibility and cost
- Detects the official WordPress AI plugin and explains how the two work together
- Anthropic prompt caching on system prompts and tool definitions reduces repeated input costs
WordPress 7 AI foundations support
PressBot 1.7.0 adds progressive WordPress 7 AI support: AI Client detection, Connector-aware setup states, workflow-focused Abilities, and a text-only public chat route through WordPress AI Client. PressBot BYOK remains the full tool workflow path for RAG, WooCommerce, bookings, lead capture, admin agent, and Trusted Plans.
WordPress 7 integration
- Detects WP AI Client, AI Connectors, Abilities API, and
wp_supports_ai()without requiring WordPress 7 - Uses WordPress Connector credentials through Core APIs for text-only public chat when available
- Registers PressBot Abilities for AI Visibility audits, Opportunity Inbox, External Knowledge search, conversation summaries, and Trusted Plan creation
- Does not import raw Connector API keys; full tool workflows continue through PressBot BYOK providers
Delete a conversation in one click — without breaking your leads
PressBot 1.6.9 adds one-session conversation deletion from the Conversations list and the Session Detail screen. Removing a session also clears the cached conversation summary, so future summaries no longer include the deleted thread. Linked leads are preserved — only the conversation reference is detached.
Cleanup that respects the rest of the system
Conversations were previously all-or-nothing. Now you can prune individual sessions — test threads, spam, or noisy QA runs — without nuking the full log or losing the leads those sessions produced.
Conversation management
- Delete a single conversation from the Conversations list or the Session Detail screen
- Cached conversation summaries are cleared on deletion so future summaries stay clean
- Lead rows are preserved — only the link to the deleted conversation is safely detached
External Knowledge answers now show source URLs
PressBot 1.6.8 makes RAG-powered answers easier to verify. When your retrieval endpoint returns source URLs, the public chatbot now renders them as markdown links so visitors can jump straight to the original transcript, video, document, or archive source.
Verification got one click shorter
The answer itself now carries the original URLs when available, so citation-style responses are not just grounded — they are easier for visitors to inspect without digging through raw retrieval payloads.
Answer verification
- Retrieved source URLs now show up directly in chatbot answers as markdown links when the retriever provides them
- Visitors can verify claims against transcripts, videos, PDFs, docs, or archive pages without leaving the answer context blind
- The External Knowledge flow stays grounded in your approved corpus instead of inventing citation targets
External Knowledge for source-grounded visitor answers
PressBot 1.6.7 adds External Knowledge, a generic retrieval connector for the public chatbot. Site owners can connect an approved RAG endpoint so visitors can ask questions against transcripts, manuals, docs, courses, policies, support material, or research libraries.
Bring your own retrieval service
Your external service owns ingestion, embeddings, ranking, and corpus cleanup. PressBot handles the WordPress chatbot layer, settings, tool routing, source normalization, UTF-8-safe streaming, and safety caps.
External Knowledge
- Added the visitor-safe
search_knowledge_corpustool for source-grounded public chatbot answers - Added settings for endpoint URL, encrypted bearer token, default scope, result limit, knowledge label, and custom retrieval instructions
- Configured external corpora are treated as approved answer domains, with guidance to cite source titles, URLs, timestamps, or document paths when available
- If matches are not useful, the chatbot should say the approved knowledge corpus does not contain enough information instead of guessing
Reliability & Safety
- Remote retrieval is bounded with caps for query length, scope length, result count, response body size, match text, source metadata, URL length, and supported protocols
- Remote snippets are stripped of HTML and normalized before reaching the model
- Public chat streaming now preserves UTF-8, including Spanish accents and other non-English characters in streamed retrieval answers
Admin Widget Polish
- Improved admin post action cards with hide controls, persistent hidden state, and cleaner status layout
- Published post cards auto-dismiss visually while preserving post context for follow-up edits
- Post cards now include refresh-page guidance so admins know when to reload the WordPress list after create, publish, discard, or update actions
WordPress 7.0 direction documented
PressBot 1.6.6 documents how the product should fit the post-WordPress 7.0 AI stack. The direction stays practical: AI Visibility, Trusted Plans, and review-first operations now, with Core AI Client, Connectors, and Abilities treated as progressive integration targets as the ecosystem settles.
Progressive AI stack, not premature claims
PressBot is positioned as the WordPress AI operations layer: audit, explain, plan, approve, apply, and report. The 7.0 APIs are integration targets, not a dependency claim.
Product Direction
- Lead with AI Visibility audits, Trusted Plans, and review-first remediation for real WordPress teams
- Describe Core AI Client, Connectors, and Abilities as progressive integration targets
- Set the product boundary that 1.7.0 now follows: progressive Core AI integration, with BYOK still powering full tool workflows
- Keep the product story anchored in safer operations, not speculative platform language
WordPress 7.0 Readiness
- Prepared messaging for the shift toward agent-addressable WordPress capabilities
- Clarified that PressBot already uses typed tools, approvals, and operator guardrails in its own admin agent
- Framed WordPress-native AI APIs as future interoperability surfaces
- Kept external-facing copy grounded in shipped workflows rather than future architecture
Copy Guardrails
- No claims that PressBot includes WordPress Core AI Client, Connector, or Ability integration yet
- No implication that every plugin automatically becomes available to external agents
- Clear distinction between current PressBot operations and future WordPress AI interoperability
- Public language stays useful for site owners and agencies, not only platform developers
AI Visibility turns gaps into reviewable plans
PressBot 1.6.5 introduces the AI Visibility Opportunity Engine, dedicated Yoast SEO and Advanced Custom Fields tools for the admin agent, and a sharper Trusted Plans review experience. Audits group answer-readiness opportunities, estimate the score lift, and turn the lowest-risk fixes into Trusted Plans you preview before anything changes.
Audit, group, plan, review — in that order
v1 writes are scoped to Yoast SEO title, meta description, focus keyphrase, the native post excerpt, and tags. Nothing changes until you apply the Trusted Plan.
Features
- AI Visibility Opportunity Engine audits pages, groups quick wins, and estimates expected score lift
- Reviewable Trusted Plans with per-action before/after diffs on pending and applied cards
- Yoast SEO admin-agent tools: get, audit, suggest, and update post meta with confirmation
- Advanced Custom Fields admin-agent tools: field groups, fields, schema, options, audits, validated updates
Improvements
- Restored, Restore failed, and Skipped status labels on Trusted Plan cards
- Apply, restore, and cancel actions redirect back to the exact card you were reviewing
- Post-apply notices show concrete success counts instead of generic confirmations
DeepSeek now streams live
PressBot 1.6.4 turns DeepSeek into a much more natural real-time chat option. Public chatbot replies now arrive token by token instead of landing as one delayed block, so the lowest-cost provider feels faster, more conversational, and much closer to the live experience visitors already get with Claude.
The budget chat option stops feeling buffered
DeepSeek now streams in the widget instead of waiting to dump the full answer at the end, which makes low-cost public chat feel noticeably faster during real visitor conversations.
Improvements
- DeepSeek public-chat replies now stream live instead of appearing as one big buffered response
- The low-cost chatbot path now feels much closer to Claude and Gemini in the frontend experience
- DeepSeek is now a more practical fit for high-volume sites that still want a responsive visitor chat experience
Bug Fixes
- DeepSeek no longer waits to dump the whole reply at once in supported hosting environments
- Public-chat follow-ups like booking checks and lead capture still stay in the same conversation flow while replies stream
DeepSeek + smarter Inbox planning
PressBot 1.6.3 expands the public-chat lineup with DeepSeek, sharpens the Opportunity Inbox into a better review layer for recurring content work, and cleans up onboarding and analytics around the full four-provider BYOK setup. Shield and Plugin Guardian also got quieter, more reliable handling in the background.
Inbox turns patterns into plans
Historical summaries now help the Opportunity Inbox spot recurring themes, raise more urgent digests, and turn stale drafts or existing-content refreshes into Trusted Plans you can review before acting.
Features
- DeepSeek V4 Flash joins the lineup as a fourth BYOK provider for the public chatbot
- Opportunity Inbox now uses historical summaries to weigh recurring themes more intelligently
- Inbox suggestions can now spin stale drafts and existing-content refreshes into Trusted Plans
Improvements
- Analytics now recognizes the full provider lineup more cleanly, including OpenAI and DeepSeek references
- Getting Started and setup guidance now show the right API-key steps for Anthropic, Gemini, OpenAI, and DeepSeek
- Higher-priority digest handling helps urgent Inbox items surface faster
Bug Fixes
- Shield now handles custom login flows more reliably without breaking the protected path
- Plugin Guardian now stays quiet during PressBot’s own updates instead of surfacing noisy review warnings
OpenAI / GPT Support
PressBot added OpenAI support alongside Anthropic Claude and Google Gemini. The current lineup also includes DeepSeek as the ultra-low-cost public chatbot option. Bring your own OpenAI API key and use the GPT-5 family for the chatbot, the admin agent, or image generation via GPT Image 2 — the same BYOK model, no markup, no middleman.
OpenAI joins the provider lineup
Paste an sk- key, pick the right GPT-5 model per surface, and mix GPT with Claude and Gemini freely across supported surfaces. DeepSeek later expanded the public-chat lineup. Image generation picks up GPT Image 2 when OpenAI is configured and organization verification is complete.
Features
- OpenAI is now a first-class BYOK provider for both the visitor chatbot and the admin agent
- GPT-5 family models available in the per-agent model picker alongside Claude and Gemini
- GPT Image 2 supported for featured-image generation and the content pipeline; Gemini image generation unchanged
- Providers can now be mixed per surface — for example Claude for the admin agent, GPT for the public chatbot, and Gemini for images
Improvements
- Setup wizard, Getting Started guide, and in-product copy updated to reflect three-provider BYOK
- Clearer error handling when OpenAI organization verification is required for image requests
Plugin Guardian & Conversation Cleanup
A Shield-layer security feature plus focused cleanup work: Plugin Guardian now audits plugin updates before and after install, and Telegram conversation history in wp-admin is cleaner and more consistent wherever admins review it.
Plugin Guardian
Snapshot plugin updates before install, flag silent ownership changes, and run AI review on the changed PHP files from Shield.
Features
- Plugin Guardian adds a Shield watchdog for plugin updates with pre-update snapshots, ownership-change detection, and AI review of changed PHP files
Bug Fixes
- Telegram-originated agent sessions no longer show duplicate messages in the Conversations page
- Internal classifier rows are now hidden from the session transcript, session list, and CSV export so only user-facing conversation content is shown
- Conversation sessions now group correctly even when internal analytics rows use a different provider behind the scenes
Improvements
- Admin conversation counts and recent-activity stats now ignore internal analytics rows so dashboard totals match the transcript you actually see
- Related conversation listings now use the same internal-row filtering for a cleaner, more consistent history view
Automations & Local Approvals
PressBot Pro moves beyond one-off chat requests. You can now schedule recurring AI tasks, route risky runs into trusted plans for review, and apply or restore changes from a dedicated Automations workspace inside WordPress.
Scheduled Automations
Run recurring AI tasks on an hourly, daily, or weekly schedule with stored or Telegram delivery and a cleaner operations-focused workflow.
Local Approvals
Manage pending plans, apply and restore history, and run smoke tests from one approvals-focused workspace in wp-admin.
Features
- Scheduled Automations for recurring AI work with hourly, twice-daily, daily, or weekly schedules
- Review-required execution mode that stages changes as trusted plans instead of applying them immediately
- Local approvals flow with pending plans, plan activity history, manual apply and cancel actions, and restore support for completed runs
- New Automations admin workspace that brings scheduled tasks and trusted plans into one dedicated screen
Improvements
- Dedicated
pressbot_agentservice user for Telegram and scheduled automations so headless actions are attributable and safer - Settings workspace reorganized into clearer Essentials, Growth, and Operations sections with a better save flow
- Shield and visitor chat runtime refined with stronger layout polish, better SSE buffering behavior, and clearer failure reporting
- Automation infrastructure now requires WordPress 6.2+, matching the SQL features used by the scheduling layer
- Billing and invoice details are now hidden from the Freemius Account page by default so license holder invoices are not exposed on agency-managed client sites
Safer Agent Execution & MCP Gating
A hardening release focused on confirmation safety, accurate MCP access control, and more reliable headless agent behavior.
Security & Safety
- Destructive admin tools now correctly trigger confirmation before create, update, or delete actions run
- MCP Pro access is now gated by the real Freemius license tier instead of code availability on staging or dev installs
- Headless Telegram and automation actions now run as a dedicated service user instead of impersonating the first administrator
Reliability
- Application-level tool failures now surface correctly in the admin agent instead of showing as false successes
- Visitor chat streaming handles buffering-prone reverse proxies more reliably
- WordPress 6.2 minimum enforced for the automation SQL layer and surfaced with clearer admin notices
The AI Interface for WordPress
PressBot becomes a platform. Your WordPress site now speaks MCP — external AI clients like Claude Code, Cursor, Codex, and other MCP agents can connect and manage your site alongside the dashboard agent, Telegram bot, and visitor chatbot. Four channels, one toolset.
MCP Server
Expose all 115 admin tools to any MCP-compatible AI client. Streamable HTTP transport, WordPress Application Password auth — no SSH required.
Knowledge Base System
Store SOPs, workflows, and business docs as a dedicated post type. Entries power the chatbot, agent, and MCP resources automatically.
Booking Management
Create manual bookings for offline clients and block vacation days — with automatic conflict detection against your calendar.
Features
- MCP Server — expose all 115 admin tools to external AI clients (Claude Code, Cursor, Codex, and other MCP agents) via the Model Context Protocol
- Streamable HTTP transport with WordPress Application Password authentication — no SSH or custom API keys required
- Knowledge base entries exposed as MCP resources — AI clients read your site context before taking action
- MCP settings panel with endpoint URL, auto-generated client config, and one-click connection testing
- 6 Knowledge Base admin tools — create, read, update, delete, and search entries via
pressbot_kbpost type with topic taxonomy. Free: 10 entries, Pro: unlimited create_manual_booking— add bookings for offline clients with automatic conflict detectioncreate_vacation_day— block calendar days with existing booking warnings
Improvements
- WP Booking Pro tools grouped in their own
bookingcategory - Knowledge base entries auto-loaded into chatbot and agent prompts for richer, site-aware responses
Booking Tools & Knowledge Base Durability
Two new admin tools for WP Booking Pro and a critical infrastructure improvement for knowledge base files.
Features
create_manual_booking— add bookings for clients who book via phone, WhatsApp, or in person, with automatic conflict detection against existing bookings, vacation days, and working hourscreate_vacation_day— block calendar days as vacation, reserved, or custom with warnings when existing bookings overlap
Improvements
- Knowledge base files moved from plugin directory to
wp-content/uploads/pressbot/— files now survive plugin updates and reinstalls - Automatic migration on upgrade with legacy-path fallback for un-migrated installs
Security Hardening & Onboarding Fix
Stronger bot user privacy, a fixed post-checkout onboarding page, and reliable conversion tracking for all new users.
Security
- Bot user fully hidden from public-facing WordPress metadata — no longer discoverable through any standard enumeration method
- Defense-in-depth: multiple layered protections for bot user privacy
Bug Fixes
- Thank-you page no longer crashes after checkout — new users now see proper onboarding
- Free download checkout flow now works correctly for all download types
Summary History, Auto Language & Tour Page
Conversation insights now persist over time, the chatbot matches each visitor's language automatically, and a new tour page lets prospects see the real admin UI before installing.
Features
- Conversation summary history — AI-generated insights are stored persistently, building a timeline of how visitor topics evolve over time
- History navigation on Conversations page — browse past summaries without re-generating them
- Sentiment trend chart on the Analytics page sidebar — see how visitor satisfaction changes day by day
- Auto language matching — the chatbot detects each visitor's language and responds in kind, no configuration needed
- Tour page — an interactive screenshot walkthrough of the admin UI, settings, agents, analytics, and more
Improvements
- Free tier stores the last 3 conversation summaries; Pro tier gets unlimited history
- Tightened focus mode — the chatbot stays strictly on-topic, no general advice or third-party recommendations
- Tour links added across the marketing site (homepage, pro page, feature pages, footer) for better discoverability
AI-Powered Setup, Smarter Writing & Better Security
Let AI handle the setup work — from conversation starters to welcome messages. The admin agent now writes posts about trending topics, and Shield gives you clearer security alerts.
Features
- AI-suggested conversation starters — click "Suggest with AI" in Settings and get 4 site-specific prompts generated from your knowledge base, ready for visitors to click
- AI-suggested welcome message — generate a personalized greeting that matches your site's tone and personality
- Smart auto-suggest — when you generate or refresh your knowledge base, conversation starters are automatically suggested if you haven't customized them yet
- Custom bot avatar — upload your own image for the chatbot bubble and header through the WordPress media library
Improvements
- Conversations table now shows what each visitor said first, so you can scan conversations at a glance without opening each one
- Conversation starters now fill the input field instead of sending immediately, giving visitors a chance to review or edit before sending
- Getting Started page is now a live checklist that tracks your actual setup progress — API key, knowledge base, widget status, and first conversation
- Admin agent can now write blog posts about current events and trending topics using real-time web research
- When the admin agent can't create a post, you'll see exactly what went wrong instead of a generic error message
Bug Fixes
- Web research in the admin agent now works correctly with Gemini models — previously returned empty results on some configurations
- Shield security alerts now correctly distinguish between real usernames and display names when reporting distributed attacks
- Checkout confirmation page is now protected against bot traffic that could trigger false conversion events
- Blog posts now show their featured image when shared on Twitter, Facebook, and other social platforms
Starter Chips, Satisfaction Ratings & KB Refresh
Give visitors a head start, find out what they think, and keep your chatbot's knowledge fresh.
Features
- Suggested prompts — clickable starter chips appear below the greeting when the chatbot opens, giving visitors quick conversation starters like "What do you do?" or "How can I contact you?"
- Customizable chips — configure up to 4 starter prompts from Settings, or use the smart defaults that work for most sites
- Conversation satisfaction ratings — visitors can rate their experience with thumbs up or down after a conversation, giving you a clear signal on chatbot quality
- Satisfaction metrics in Conversations — see your overall satisfaction percentage at a glance, plus per-session ratings in the table and session detail views
- Satisfaction data in CSV exports — ratings are included when you export conversation history
- One-click knowledge base refresh — update your chatbot's knowledge when your site content changes, without regenerating from scratch
Auto-Generate Knowledge Base
Your chatbot should know about your site from the start. Now it can.
Features
- One-click knowledge base generation — PressBot reads your pages, posts, and products, then creates a ready-to-use knowledge file so the chatbot can answer visitor questions immediately
- New setup wizard step — after connecting your API key, the wizard offers to generate your knowledge base before you go live
- Generate button in Settings — create or regenerate your knowledge base anytime from the Knowledge Base sidebar
Improvements
- Stronger chatbot boundaries — the visitor chatbot now gives clearer responses when asked to do things outside its scope, like creating pages or changing settings
Activation & Onboarding Fixes
Smoother first-run experience for new free installations.
Bug Fixes
- Fixed an issue where new users could see a permissions error during first-time plugin activation
- Fixed welcome overlay appearing too early — it now waits until the initial setup is complete
- Fixed download links in welcome emails pointing to the wrong location
Improvements
- Improved onboarding flow so each step completes before the next one begins
Initial Public Release
The first public release of PressBot Pro — an AI-powered WordPress assistant with a visitor chatbot and admin agent.
Core Platform
- Visitor chatbot widget with knowledge base integration, dark mode support, and WooCommerce product search
- Admin AI agent with 115 tools across 20 categories for full site management through conversation
- BYOK model — originally Anthropic (Claude) and Google (Gemini) API keys, with later releases adding OpenAI and DeepSeek. No per-message fees, full control over costs and privacy
- Per-agent model selection — choose different models for the chatbot vs the admin agent
Content & Media
- 8-stage content pipeline — research, strategize, write, critique, generate images, categorize, and publish
- AI image generation via Gemini with multiple aspect ratios and automatic featured image assignment
- AI vision — auto-generate alt text, captions, and descriptions for media library images
- Bulk operations with real-time progress reporting — process up to 100 posts or 50 images at once
Integrations
- Telegram bot — manage your WordPress site from your phone with the same 115 tools
- WooCommerce — 10 dedicated tools for products, orders, coupons, and sales analytics
- Google Analytics 4 — query real traffic data through the agent, composite site reports
- WP Booking Pro — visitor availability checks through conversation
Security & Intelligence
- PressBot Shield — AI-powered login protection with auto-lockouts, threat analysis, IP blocking, and custom login URLs
- 12-check security audit covering file permissions, debug mode, SSL, admin username, and more
- Agent memory — persistent preferences and context across sessions, auto-saved proactively
- Web search integration for real-time research and current information
Engagement & Customization
- Proactive messaging — exit intent and timed nudge triggers with configurable messages and suggestion chips
- Lead capture from chatbot conversations
- Explain CTAs — add contextual "Ask PressBot" buttons anywhere via data attributes, shortcodes, or JavaScript API
- Contextual suggestion chips that change based on the current admin screen
- Full conversation logging with export to CSV, message statistics, and configurable retention
Stay up to date
New version, same idea.
PressBot Pro users receive automatic updates through Freemius. New here? Try PressBot for free.