产品发布当前 v1.0.2

更新日志

清晰记录 Namiru.ai 的每项变化，从新功能到让日常工作更顺畅的小修复。

新增

The homepage now features a transparent, silent mascot animation with a static poster fallback.
The homepage now highlights answers, bookings, lead capture, and conversation insights across refreshed localized copy.
How It Works now shows an animated crawl proof and clearer business learning, installation, and reporting steps.
The homepage now presents six use cases in a static grid and includes Bookings in the conversation insights preview.
The Why Namiru section now uses an editorial stat band with pricing, message, availability, and Reddit proof.
Homepage platform guidance now links directly to the widget framework quick-start documentation.
Plan prices, allowances, annual savings, and top-ups now share one configuration source across product surfaces.
Approved Autopilot prospects now show their send time, status, Autopilot badge, and shared review dialog in Prospects.
Outreach lead review now provides a canonical country picker covering all supported text-only markets.
Discovery runs now expose durable stages, elapsed time, model details, counters, outcomes, and recovery status.
Autopilot now continues searching for usable leads until its target is met or bounded stop conditions are reached.

变更

The mascot video loads only after its poster, plays while mostly visible, and pauses in background tabs.
Visitors who prefer reduced motion receive the static mascot poster without downloading the video.
Homepage languages now prioritize key European markets before the wider global language set.
The homepage buyer FAQ now focuses on WordPress, Shopify, and Wix compatibility.
Homepage copy now distinguishes current Shopify script support from the future dedicated Shopify plugin.
Pricing, analytics, backend enforcement, prompts, localized copy, and emails now consume shared plan values.
Backend container builds now include shared plan configuration, and deployment detection tracks changes to it.
Autopilot keeps unapproved discoveries in its review flow until approval creates a real prospect.
Unavailable and email-less discoveries remain visible in All but are excluded from actionable review queues.
Production application rollouts now run sequentially as backend API, backend AI, then frontend and stop on failure.
Test is now the default remote deployment target for automation and deployment menus.
Calendar availability now requires at least one valid published window, while empty or closed schedules disable booking.
The live demo tutorial now highlights the complete widget invitation and advances to the signup spotlight.

修复

Long-running page visits no longer cause state synchronization failures when their duration exceeds database limits.
Invalid, negative, fractional, non-finite, and non-numeric analytics durations now return a client error before persistence.
Selecting Slovakia in outreach review now stores its supported SK country code instead of an autofilled invalid value.
Linked Autopilot prospects no longer expose a manual composer that could bypass reviewed email copy.
Prospects and Autopilot now use one canonical lead representation, keeping linked lead details consistent.
Active discovery requests now reconnect to the existing run instead of failing with a conflict.
Discovery workers now renew ownership and mark abandoned runs interrupted for reliable recovery.
Outreach language choice now applies consistently through discovery, preview, approval, retry, immediate send, and scheduling.
Language-mismatched personal notes now fall back to reviewed static outreach copy.
Leads now tolerate older API responses that omit summary data during rolling deployment overlap.

安全

Production deployment entry points now require approval from Patrik in a local interactive terminal before any external mutation.
Production gates now reject automation markers, agent process ancestry, non-interactive callers, and invalid approval handoffs.
Public package publishing now requires the same human-only production approval before version or registry changes.
Production approval is passed to selected deployment scripts through a short-lived, restricted capability that is cleaned up on exit.
Outreach site scanning validates every redirect, limits streamed response size, and pins requests to validated IPv4 targets.

v1.0.1

2026年8月18日

新增

Customers can choose bookable services, durations, weekly hours, multi-day appointments, labels, and automatic time zone handling.
The chat widget now shows service cards and an interactive calendar slot picker.
Customers can cancel bookings through secure cancellation links.
Booking and changelog content is available across all 15 supported locales.
Widget documentation now creates personalized AI installation prompts for Codex, Claude, Kimi, and Muse.
AI-assisted installation is now the first embed option in Agent Builder and the creation wizard.
A changelog page and localized changelog content are now available.
Admins can review AI spending by user, IP, URL, model, source, or day.
The AI spending dashboard separates chat and crawl costs, tokens, events, exact costs, estimates, and unpriced usage.
Admins can filter, search, sort, paginate, chart, and export the current AI spending view as CSV.
Registered users now display a country flag and attributed AI spend with a link to their filtered ledger.
Admins can configure operational alerts and review subscription, critical error, test, and health-check events in Alert Center.
New paid subscriptions and critical API or AI failures can notify configured operator email recipients.
A daily production AI health check verifies agent initialization, an exact sentinel response, and completed streaming.
Admins can view the latest AI health-check status and filter health events in Alert Center.
Daily backups now protect paid-user settings, billing data, and files on persistent storage.
Admins can run, inspect, expand, download, and delete backups from the Backups page.
Backups use change detection, retain two successful archives per user, and expose the latest seven run records.
Admins can manage outreach prospects, discovery runs, review queues, per-country settings, and automated sending.
Outreach discovery supports model, provider, and reasoning-effort selection with durable run snapshots.
Outreach discovery backfills unusable candidates until the configured usable-lead quota is reached.
Admins can preview outreach snapshots and verify whether leads have embedded the widget.
AI usage records now capture provider, model, source, tokens, cost confidence, IP, country, and request URL.
Temporary agents and public crawls now retain anonymous AI usage attribution by agent, IP, URL, and session.
Bookings now store customer time zones and service snapshots for reliable historical context.
Lead enrichment now extracts normalized phone numbers and date details.
The WordPress plugin now includes updated settings, publishing guidance, documentation, cleanup on uninstall, and a packaged release archive.
A live widget preview and expanded framework and WordPress test pages are now available.
Release preparation, validation, and release-note tooling now provide a repeatable release workflow.
An admin spending API exposes summaries, grouped rows, daily timelines, model options, and source breakdowns.
Backup run records now include per-user outcomes, failure categories, archive counters, and expandable details.

变更

Blog generation, blog SEO, Midjourney prompt creation, and support summaries now use Luna with xhigh reasoning.
The Midjourney prompt helper now uses an authenticated backend request instead of calling Cerebras from the browser.
Outreach now defaults to Luna with xhigh reasoning while preserving compatible model and provider overrides.
Public widget AI usage is attributed to the configured agent owner rather than the optional widget viewer.
All model calls in a chat graph invocation, including tool selection and final answers, are accumulated for usage reporting.
Conversation-enrichment calls and retries are now included in chat usage costs.
Exact provider costs take precedence, while missing costs use labeled estimates or remain explicitly unpriced.
The Registered Users table now shows compact plan and status badges without verbose billing subtext.
Recent payments now use dense rows, truncate long text, show two entries, and summarize additional events.
Recent Payments now contains payment-like events only, while audit and lifecycle events remain in audit storage.
Long registered-user names and emails now use ellipsis with full values available on hover.
Paid feature controls now prefer backend entitlements and retain legacy plan fallbacks when needed.
Custom color controls now state that a Starter plan is required.
The AI installation command and manual snippets update together when a user selects another bot.
Logged-out widget-doc visitors receive login and signup actions that return them to the documentation.
Logged-in users without a bot receive a direct create-bot action, while loading failures offer Agent Builder recovery.
Automatic backups now use a persisted rolling 24-hour cadence that survives restarts.
Manual backup runs no longer move the automatic daily backup schedule.
Backup archive retention is fixed at two successful archives per user and run history is bounded.
Backup start requests return immediately for polling, avoiding long API requests during archive creation.
An all-users backup with no eligible users now completes successfully with zero users scanned.
Operational event retention is configurable from 1 to 60 days and cleanup runs at startup and hourly.
Critical alert emails now use a full-width incident layout with structured metadata and escaped stack evidence.
Critical error emails use environment-specific senders, and test-environment subjects are clearly tagged.
Outreach discovery reasoning effort now follows the selected model's advertised options.
Agent configuration caches now include ownership and are invalidated when a temporary agent is claimed.
The widget can now initialize automatically from a script tag containing an agent ID.
Backup deployment now applies required persistent-volume claims before the application deployment.
Booking notification emails and other transactional email templates received refreshed presentation.
Homepage, onboarding, pricing, FAQ, footer, theme, use-case, and responsive preview content received broad updates.
Offscreen homepage animations now pause to reduce unnecessary rendering work.
The admin funnel dashboard now uses shared funnel-stage definitions and expanded session analytics.
Backup archive creation now streams directly through compression and size enforcement to its destination.

修复

Public widget messages without a viewer token are no longer omitted from AI usage reporting.
Anonymous temporary-agent chats are now persisted instead of disappearing from the spending ledger.
Crawl responses without provider usage metadata now use token fallback instead of being omitted.
Partial or blank streamed chat responses now preserve usage metadata already returned by the provider.
Claiming a temporary agent now clears stale cached ownership so later usage is assigned correctly.
Billing updates with partial or legacy payloads now preserve existing user and billing fields.
Disabled or downgraded accounts no longer appear to have active payment grace from stale dates.
Long payment descriptions no longer widen or overlap the Registered Users table.
Duplicate Stripe webhook deliveries no longer send duplicate new-subscription emails.
Daily health checks now handle explicit initialization failure, exact response validation, stale claims, and competing finalizers.
Operational alert email fallback now proceeds when event deduplication storage is unavailable.
Missing backup artifacts are detected, marked failed, and recreated instead of being reported as unchanged.
Backup pruning now preserves a per-user archive floor and uses actual disk sizes for capacity calculations.
Failed artifact deletion now preserves its database record so reconciliation can recover safely.
Backup storage listings no longer mass-demote records when the configured root is missing or empty.
Backup migrations now add failure fields safely to existing installations and backfill prior failures.
Backup streaming now propagates pipeline errors and includes partial staging files in capacity accounting.
Calendar availability now accounts for Google Calendar busy intervals and configured calendar permissions.
Calendar tool history, flexible inputs, multiple durations, and widget slot selection received regression coverage.
Phone numbers are now normalized consistently across AI enrichment and API processing.
Conversation topic enrichment now records retries and preserves broader contact information.
The outreach pipeline now retains unavailable, email-missing, and iframe-blocked candidates in the complete list.
Admin plan changes no longer place audit-only events in the Recent Payments payload.
Booking coverage now distinguishes downloadable successful archives from failed records that retain forensic paths.

安全

Backup archives now redact tool headers, query parameters, request bodies, data-feed credentials, URL user info, and booking cancellation tokens.
OAuth connection and refresh-token records are excluded from backup archives.
Backup manifests now identify nine redacted field paths for operator verification.
Operational alerts redact authorization data, cookies, passwords, secrets, tokens, and API keys before storage.
Global error hooks no longer capture request bodies, query strings, or authorization headers.
Local operational email delivery is blocked both during event reporting and at the final email sender.
Daily AI health checks run only on approved production hosts and have no public execution endpoint.
A database uniqueness claim prevents multiple pods from sending duplicate daily health-check conversations.
Health-check status and operational alert controls require administrator authentication.
AI spending, backup, outreach, and service-management administration routes require authenticated access.
The browser no longer holds credentials for Midjourney prompt generation.
Admin spending queries use validated group and sort allowlists with parameterized search and dates.
Backup downloads require a successful archive state instead of trusting the presence of a stored file path.
Environment example files replace checked-in backend environment files, reducing accidental secret exposure.
Backup deletion and reconciliation now avoid creating unrecoverable orphaned artifacts or misleading coverage state.

v1.0.0

2026年8月13日

新增

Created AI support agents by crawling a website and turning its pages into an editable knowledge base.
Added AI-powered answers that use each business's own content and respond in the visitor's language.
Added a guided agent builder for configuring instructions, behavior, knowledge, appearance, and integrations.
Shipped a plug-and-play chat widget in floating-button and inline modes with isolated styles for reliable embedding.
Added one-line script installation plus official WordPress, React, Vue, and Angular integrations.
Added custom colors, fonts, logos, avatars, greetings, launcher styles, and optional Namiru branding removal.
Added pre-chat questions and quick replies that help route visitors before a conversation begins.
Added persistent visitor conversations, new-chat controls, transcript delivery, and conversation feedback.
Added lead capture and human handoff flows for visitors who need personal help or show buying intent.
Added an Insight Hub with AI-generated summaries, topics, pain points, lead signals, and searchable conversation history.
Added bookable services, weekly availability, time-off rules, booking management, and Google Calendar synchronization.
Added in-chat service selection and appointment booking so visitors can move from a question to a confirmed time.
Added a customer dashboard for managing multiple agents, domains, conversations, leads, calendars, and usage.
Added free, Starter, and Pro plans with monthly or annual billing, conversation allowances, and optional top-ups.
Added a localized marketing site, guided product trial, documentation, blog, contact pages, and public changelog.
Added an operator panel for user management, content, funnel analytics, outreach, AI spend, alerts, logs, and backups.

变更

Set the free plan to 20 AI conversations per month with one agent, one domain, analytics, and no credit card requirement.
Made the WordPress plugin load the current Namiru widget from the CDN so improvements arrive without plugin updates.
Expanded calendar setup from one generic appointment type to individually managed services with names, prices, and durations.
Standardized production releases on Semantic Versioning with generated notes, retry-safe deployment, and public release history.

修复

Improved website crawling recovery, progress reporting, and knowledge review when a session expires or a crawl stalls.
Kept development chat pages on development services so local testing no longer creates production conversations.
Strengthened booking availability around business time zones, daylight-saving changes, overlapping slots, and closed schedules.
Improved billing state handling so failed payments, grace periods, downgrades, and plan entitlements stay consistent.

安全

Hosted customer data in the European Union and applied privacy-focused defaults across the widget and dashboard.
Protected public booking actions with scoped cancellation tokens and prevented customer lookup data from exposing them.
Added signed Google Calendar connection state and server-side validation for calendar authorization callbacks.
Added rate limits, input validation, safe content rendering, authenticated admin routes, and operational failure alerts.