Skip to main content
A running record of significant updates to the Profy platform. We iterate every week — here are the changes with the greatest impact on user experience.

​
July 2026

The all-new standalone developer console platform.profy.cn is now live, aligned with OpenAI Platform. Manage API Keys, view call volume trends and logs, and debug endpoints in the Playground — all in one dedicated console.Simultaneously released are unified Python and TypeScript SDKs:
API routes are unified under the /v1/* namespace (/v1/agents/run, /v1/chat/completions, /v1/models), aligned with the OpenAI interface style.
Complete the entire Expert lifecycle — distillation, configuration, publishing, and earnings review — directly within your IDE. Built on an MCP remote service (22 tools), Codex and Cursor users can operate the full Expert lifecycle from the editor sidebar after installing the plugin, without switching to a browser.A distillation progress Widget is embedded in the IDE side panel, displaying 12 status fields and 6-stage progress in real time.
Expert memory has been upgraded from linear accumulation to a temporal knowledge graph. The system automatically distills core memories from conversations (entities + facts + relationships), tracking how knowledge evolves over time — contradictions don’t delete history; old facts automatically close their validity period and new facts take over the current state.Three new tabs for intuitive viewing:
  • Memory List: Full CRUD, with enable/disable for individual memories
  • Core Memories: AI-distilled entity-relationship graph
  • Memory Graph: Visualized knowledge evolution trajectory
Retrieval goes through candidate screening → graph expansion → re-ranking, capped at 20 entries / 2,000 tokens to ensure conversation quality doesn’t degrade as memory volume grows.
You can now see exactly which memories the Expert used in each conversation turn — memory sources are displayed as footnotes at the bottom of responses, clickable to view source snapshots.New “Temporary Chat” mode: when enabled, the current conversation does not read historical memories, does not write new memories, and leaves no traces whatsoever. Ideal for testing, sensitive topics, or one-time consultations.
A new plan mode aligned with Cursor: once activated, the Expert only researches and plans, maintaining an iteratively editable plan document. You refine the plan content through conversation, and only click the Build button to start execution when satisfied.The plan document is persisted in the sandbox, retained across turns. In plan mode, the Expert won’t execute any write operations — this isn’t an approval dialog, but a true “think it through before acting” mode.
Creator Studio now features a unified analytics page, replacing the previously scattered feedback/revenue/insights panels. View everything on one page:
  • Trend charts: Views, conversations, purchases by day/week/month
  • Source analysis: Where traffic comes from (marketplace, share links, promo notes, search)
  • User profiles: Professional distribution of users engaging with your Expert
  • Revenue details: Revenue trends and conversion funnels
The Team plan has been upgraded from “submit a lead form” to online self-service purchase. Select the Team plan on the subscription page → enter a team name → pay via WeChat/Alipay → the organization is automatically created with a shared credit pool injected.Supports monthly and annual billing tiers, with seats and pool amounts configured in real time by operations in the admin dashboard. One person per organization — users already in an organization will be prompted.
Aligned with Cursor’s multi-identity model: you can now hold both a personal plan and team membership simultaneously. Switch your “active subject” from the avatar in the bottom-left corner — choose to consume as a personal identity (charges personal wallet) or as a team identity (charges team credit pool).Purchasing a Team plan no longer requires canceling your personal subscription first. After purchase, you’re automatically switched to the corresponding identity, with plan badges and credit balance syncing in real time.
When a Team plan expires, it enters a frozen state: consumption is blocked, but the organization structure and member relationships are fully preserved. Renewal instantly restores everything — no need to re-invite members.The Owner can switch to their personal plan at any time to continue using the platform. The system automatically sends the owner a one-time in-app notification upon expiry.
The all-new Plugin Marketplace ecosystem is live, 1:1 aligned with Codex:
  • Browse & install: Discover community plugins in the marketplace, install to your Expert with one click
  • Creator publishing: Publish your own plugins through a dual-track review mechanism
  • Skills & MCP: Community plugins carry skills and MCP servers declaratively
  • Seamless integration: Installed plugins are automatically loaded during conversations, with remote MCP isolated by namespace
Have an Expert help you build a website during a conversation. Select the Sites plugin and the right panel automatically opens a preview window (with HMR live refresh); the Expert writes code in the sandbox, and you deploy to EdgeOne with one click after confirming.Built on the TanStack Start framework with SSR support. After deployment, you get an independent URL — no need to manage your own servers.
The right-side workspace has been split into three independent tabs:
  • Preview: Dev server live preview (auto-refreshes while coding)
  • Widget: MCP Apps interactive components (sandboxed iframe isolation)
  • Browser: Remote browser (noVNC, with Chrome-style toolbar)
The three panels are independent, non-interfering, and open automatically as needed.
The knowledge base has been rebuilt from a self-hosted RAG to external platform connectors. No more re-uploading documents — directly connect your:
  • Feishu docs and knowledge bases
  • DingTalk documents
  • Yuque knowledge bases
  • IMA knowledge platform
Authorize once in settings, and your Expert can search and reference your external knowledge in real time during conversations.
Platform administrators can now create custom roles by selecting permissions from a permission tree organized by module, then assign them to team members. Aligned with Alibaba Cloud RAM’s permission management model.11 modules × 18 permission codes, with fine-grained control over “who can see what, who can change what.” System preset roles are read-only protected, and deleting a custom role requires checking whether any members are currently using it.
Operations can now issue free gift cards directly from the /admin/gift-issue dashboard without going through the paid purchase flow. Supports batch card generation for marketing campaigns, partner gifting, or customer service compensation.User claiming and redemption reuse the existing gift card flow, and operations can void unclaimed cards in bulk.
Profy Desktop adopts a new 2-process stdio JSON-RPC architecture (aligned with OpenAI Codex). Electron communicates with the Python sidecar through stdio pipes:
  • Zero TCP port exposure, significantly improved security
  • Tool execution fully covered by Seatbelt sandbox
  • Supports CUA (Computer Use Agent) and browser automation
Desktop provides the complete Profy experience, ideal for scenarios requiring local file operations and system-level tools.

​
June 2026

The all-new “Content Plaza” is live — an inspiration showcase platform contributed to by creators and users alike. The waterfall feed supports four media types:
  • Images: Multi-image carousel support
  • Videos: Immersive playback
  • Music: Record-style player + LRC lyrics sync
  • Interactive HTML: Embedded interactive games and visualizations
Supports likes, comments, favorites, and following creators. See a piece you like? Click “Make similar” to directly start a conversation with the Expert to recreate it.
Any logged-in user can publish promo notes to share their creations made with Experts. Supports AI-generated copy and cover images (generated asynchronously, not blocking the publish flow).Notes are linked to Experts, helping potential users discover and understand what an Expert can actually do. Publishing goes through a dual-track review mechanism — when editing an already published note, the old version stays online while the new version is reviewed independently.
When an AI’s single response exceeds the model’s output token limit, the system automatically detects the truncation and silently continues writing up to 3 times — completely transparent to you, appearing as one seamless response.After recovery attempts are exhausted, an amber-colored tag appears with a “Continue generating” button for manual continuation. The mechanism is aligned with Claude Code’s withholding pattern.
The marketing site now includes a full blog system at profy.cn/blog. Operations write articles using a Markdown editor in the admin panel, and posts go live within seconds (ISR + on-demand revalidation, no full site rebuild required).Supports article categories, featured/pinned articles, reading time estimates, and automatic SEO optimization. The marketing site has been simultaneously upgraded to SSR/ISR dynamic rendering, so blog content is no longer blocked by static builds.
Subscription plan pricing and numerical benefits (credit amounts, storage quota, etc.) have been migrated from hardcoded values to real-time ops-configurable settings. Adjust prices in the “Platform Configuration” tab at /admin/billing, and changes take effect platform-wide within 60 seconds — no release required.Signup bonus credits are also configurable. The marketing site pricing page, subscription page, and payment page all read from the live catalog end-to-end.
Text-to-speech (TTS) has been changed from per-call to per-character billing, and speech recognition (ASR) to per-duration billing. Billing granularity now aligns with upstream providers — long text/audio no longer costs a flat per-call fee.Two new billing units have been added: per_ten_thousand_characters and duration_second, with rates precisely configurable in the admin panel.
Experts in Feishu group chats can now operate Feishu documents and spreadsheets under your identity. Send the /auth command in a group chat and scan the QR code for a one-time authorization.After authorization, documents created by the Expert belong to you (not the bot), so you can edit them directly in Feishu. Tokens refresh automatically on expiry, with re-authorization prompted when they become invalid.
Referral reward amounts, inviter caps, and counting rules are all now self-configurable by operations in the admin panel. Supports different invite caps per plan tier (Free: 10, Pro: 15, etc.) — exceeding the cap only stops inviter rewards while invitee rewards remain unaffected.The invite popup displays the current plan’s invite cap in real time and guides users to upgrade their plan for more slots.
When using an Expert’s scheduled task feature in DingTalk channels, task execution results are automatically pushed as DingTalk messages — no need to manually check back on the web.