سجل التغييرات
كل تحديث لـ PromptOps Manager، من أول إصدار حتى اليوم.
تنزيل PromptOps Desktop
Resolves a native binary issue that crashed the app right after auto-updating on Windows ("not a valid Win32 application"). The Windows package now ships the correct win32-x64 binaries instead of darwin-arm64. Users stuck on 0.9.203 need to uninstall and manually reinstall 0.9.205 from the website (auto-update can't kick in if the app fails to start).
The download page on promptops.it / promptoperations.ai now shows dedicated buttons for Mac Intel (alongside Apple Silicon) and Linux x64 (alongside ARM64). The backend exposes assets by architecture via API (platforms_by_arch); the frontend surfaces them automatically based on browser detection.
New modal to enter a referral code AFTER account creation, useful for users who didn't have the link at first signup or got it later. Referral tracking is now resilient to the registration sequence and entry channel.
User info now includes progress toward unlocking PRO Life based on the number of active referrals. The bar is visible from the profile and referral dashboards, giving clear feedback on the gamification path and motivating new invites.
Added a user preference to tune the sox fallback window (1–6 seconds) and a tap-to-toggle mode as an alternative to push-to-talk. More control for users who use voice dictation in different patterns depending on context.
Support for passphrase-protected SSH keys via askpass helper integrated into the git-setup flow. No more need to manually run ssh-add before cloning private repositories: PromptOps handles the prompt directly.
Every authenticated API request now sends X-Platform and X-App-Version headers, enabling user segmentation by OS and version, targeted troubleshooting, and per-release rollout statistics.
New admin view with top inviter leaderboard, presence filter (online/offline), and PRO status sort. Unique nationalities count exposed in the global ranking, prompt count consistency fix between profile and leaderboard.
Complete refactor of the desktop deploy script with pre-flight asset checks before pushing, streaming upload + automatic retry on transient errors (EPIPE/ECONNRESET/timeout), and a final 15/15 completeness check against the reference. No more half-published releases in production: if even a single file is missing, abort before touching GitHub.
System-wide global shortcut (⌘+Shift+Space) to trigger voice dictation from anywhere on the desktop, even when the app is not focused. Choice keeps macOS Spotlight free. Explicit opt-in from the bubble guide: the shortcut is not registered until the user enables it, avoiding silent conflicts with other tools.
New centralized voice intent router that dispatches every transcribed command to the right action in the webapp: focus a session, send a prompt, switch sub-agent, open panels. Extensible voice intent registry with declarative patterns and priorities, with focus/send resolution based on the current renderer state and i18n feedback.
Voice command patterns extended to Spanish, French, and Portuguese, with full i18n coverage of the most frequent actions. Command recognition is independent from interface language: speak the command as it comes naturally, the router understands and maps it.
Support for two-step composite voice commands: "open session" → "Build Pipeline". The router keeps intermediate state for a few seconds, allowing you to complete the intent without repeating the full command. Useful for disambiguating commands requiring a textual parameter.
New dedicated section in user preferences to configure push-to-talk: enable/disable, change the key combination, and conflict diagnostics. Preferences persisted Electron-side via voice-shortcut-config-store and applied live without restart.
When you manually select a sub-agent from the UI, the voice focus is automatically mirrored to the corresponding terminal. Speak a command without having to re-aim your voice first: the active context always follows your visual selection.
Added a backend proxy endpoint for audio transcription via Groq Whisper, integrated into the transcription provider fallback chain. Lower latency than the standard Whisper API, with server-side cost capping and no API key exposed on the client.
Sox fallback recording window compacted from 6 to 2 seconds to better match the push-to-talk pattern: speak, release, command fires. More responsive transcriptions and less useless audio sent to the APIs.
After every desktop update a modal appears with the new version highlights, pulled directly from the website changelog. Users see what changed right away without having to hunt for release notes. The "seen" flag is persisted server-side and reset for all users when the admin publishes new notes from the panel.
Every newsletter signup tracks the referrer: whoever brought in a new user sees their score appear in the referral dashboard. Eager-loaded referrer in all relevant queries to avoid N+1 and hardcoded referral download URL to guarantee stable links in transactional emails.
The account activation flow now automatically verifies the user's email: no more double "subscribe + confirm" step. Polished password page layout with better visual hierarchy and clearer error messages.
Every login and every authenticated API request updates a last_seen_at field on the user. Foundation for engagement dashboards, retention recomputation, and segmentation of dormant users for re-engagement campaigns.
The presence heartbeat tracks the last platform used by each user (macOS/Windows/Linux). Enables OS adoption analysis and targeted support when platform-specific bugs surface.
All email notifications now respect the user's email_notifications_enabled flag. Users who disabled emails no longer receive task assignments, comments, or team notifications — only critical transactional communications stay active.
When you reply to feedback from the admin panel, the feedback is reloaded from the server to sync state. No more inconsistent states between local UI and backend after a reply.
Prompts created in a session terminal are now correctly attributed to the session's time window for ranking calculation. More accurate counts and better fairness in the global leaderboard.
Native integration of OpenCode, the open-source local-first coding CLI. Dedicated adapter that spawns the opencode command directly with project context materialization, session tracking and shared prompt library. Seventh option among the AI providers managed by PromptOps, alongside Claude, Codex, Gemini, Copilot, Cursor and Shell.
Unified cross-provider consumption dashboard inside the desktop. Tokens, estimated costs and plan quotas per connected provider, with breakdown by session, sub-agent and time window. Designed to keep Claude Pro/Max under control alongside pay-as-you-go APIs.
Real-time health monitoring of every session terminal: hang detection, zombie processes, stalled output, CPU spikes. Health indicator in the sidebar and session view, with quick actions for clean restart without losing session state.
Step-by-step onboarding flow for new users: provider connection, first session creation, workspace selection, key features tour. Server-side completion tracking to personalize follow-up suggestions and unlock advanced content once basics are mastered.
Fully redesigned composer for spawning new sessions: provider selection with preview, inline workspace picker, prompt library accessible directly from the composer, contextual sub-agent suggestions. Fewer clicks to start, more clarity on what you are about to launch.
Each user can now set a preferred AI provider at the account level, used as default on new spawns and composer suggestions. Choice synchronized between desktop and webapp, with fallback to the available provider if not configured.
Layout, open panels, active tab and terminal splits persist per session. When you return to a session you find it exactly as you left it. Explicit persistence: view_state is saved only when provided, avoiding accidental overrides.
Cross-provider shared context layer: files, snippets and notes injected into the preamble of Claude, Codex, Gemini, Copilot, Cursor and OpenCode consistently. Each provider receives the context in the format it understands best, while keeping the same source of truth.
Real-time events when provider usage probes fail (rate limit, expired auth, exhausted quota). In-app notification with suggested recovery action and tracked event for provider reliability analysis.
Searchable combobox for nationality selection in the user leaderboard, with multi-language support. Fuzzy search that also works when typing the country name in your own language.
When a user is banned the realtime session is closed immediately on the first rejected presence flush. No more zombie connections or residual access to private team channels.
User hour tracking stats flow into the growth/admin dashboard. Aggregated view of logged hours per project, user and period — foundation for adoption and efficiency analysis.
Token-saving mode with three intensity levels: Lite (concise), Full (telegraphic), and Ultra (maximum compression). Global toggle from the bottom bar with per-session override. The eco instruction is injected into the active terminal and sub-agent preamble, forcing compressed responses that reduce output token consumption. State persisted in localStorage with automatic restore on restart.
The user leaderboard switches from tokens saved to number of prompts created as the primary metric. Explicit opt-in to participate in the public leaderboard. Server-side pagination including all opted-in users. Sort column validation and whitelist for security.
Introduction of CSS design tokens for colors, spacing, typography, and border-radius throughout the webapp. Removal of AI-generated theming anti-patterns and restoration of the indigo/purple brand palette. Full tokenization of the topbar, sidebar, session-view, action-center, and all modals.
Fully redesigned sidebar: higher density, improved contrast, compact layout. Bottom bar rethemed with an indigo hue. Auto-approve integrated with a quick-switch for the terminal directly in the sidebar.
Graphical visualization of project structure with automatic context re-injection. Navigable map of relationships between files, modules, and dependencies. Structural context is injected into sessions to orient the agent without blind exploration.
Guided assistant for recovering from push errors. Automatic problem diagnosis (authentication, remote, branch) with step-by-step resolution suggestions. Integrated into the desktop session git flow.
Granular terminal selection for auto-approve. Each terminal panel has its own independent switch. Quick-switch integrated in the sidebar for fast activation without navigating to settings. Hardened toggle against devtools tampering.
Tokens for animations and transitions (duration, easing) applied globally. Redesigned empty states with illustrations, polished standalone terminal. Intent modal plan review with redesigned layout and accessible-contrast sidebar badges.
User hour tracking with support for manual entries and automatic system recordings. Per-session time log with timestamps and metadata. Foundation for time-spent reporting by project and task.
File upload limit increased to 20MB. Backend configuration updated with aligned client-side validation for task attachments and project documents.
Automatic generation of structured action plans from user intent. Dedicated modal with preview, editing, and step-by-step approval. Intent service with multi-step planning logic and automatic prompt enhancement that enriches the request with project context. Typed IPC handlers for spawning the plan directly from the desktop.
Comprehensive knowledge base metrics per session: indexed files, executed queries, context retrieval hit rate. New KnowledgeEngineStat model with persistent tracking and a dashboard API integrated with the Token Efficiency Engine to calculate savings derived from the engine.
Public global leaderboard of users ranked by tokens saved. RankingController endpoint with pagination, time filters, and breakdown by optimization method. Token-saving gamification to incentivize use of efficiency features.
Auto-approve mode now configurable per individual session instead of globally only. Each session maintains its own persistent toggle with a real-time event on change. Enables unattended flows on specific sessions while keeping manual control on others.
Efficient terminal output streaming for live follow without overhead. The SessionTailer intercepts only deltas from the point of connection, reducing IPC traffic and CPU load on long-running sessions. Useful for monitoring background sessions or multiple views on the same terminal.
Complete account deletion flow for SSO users (Google/GitHub). AccountDeletionService with provider verification, soft delete of user data, and automatic cleanup of associated S3 assets (avatars, task documents). Privacy settings extended with granular GDPR-compliant controls.
Automatic digital signing of Windows builds integrated into the deploy pipeline. sign-win.js script with EV certificate and trusted signer support. Removes the SmartScreen warning on installation, improving perceived trust for Windows users.
Full Portuguese (pt_BR) support across the webapp, backend, transactional emails, and marketing site. Over 1,800 strings translated and integrated into the i18n system. Language selector in the topbar with user-level preference persistence.
Admin user panel extended with segmented filters: account status (active/banned), subscription tier (pro/free), authentication provider, notification preferences, platform usage (macOS/Win/Linux), and prompt creation activity. Foundation for cohort analysis, targeted campaigns, and user segmentation.
Backend middleware that records the installed desktop version for each authenticated API request. Analytics on version distribution across users and detection of users on outdated builds. Foundation for targeted in-app update notifications.
Resolved several session edge cases: session name as unique key (replaces workspace_name), automatic restore of soft-deleted sessions on re-create (avoids unique constraint violations), enforcement of soft delete with query scope, fallback and backfill for session-workspace linking on orphaned sessions.
Pro promo for 90 days activated for social users (Google/GitHub) on first login, with automatic role assignment at user creation. Promo extended to returning users re-entering after expiry. Simplified post-auth redirect by using the URL provided directly by the provider.
"Prompts created" column in the admin user table with sort support. Fixed count that was being lost during pagination — user metrics now remain consistent across page changes and filter applications.
Voice commands integrated into the session. Cross-platform audio recording (MediaRecorder in the renderer or sox in the main process). Multi-fallback transcription chain: Google Speech API → native macOS SFSpeechRecognizer → OpenAI Whisper API → local Whisper CLI. Support for 16+ languages with automatic detection.
Mode that automatically approves all pending interactions (confirmations, actions, inputs). Global per-session toggle with a real-time auto-approve:changed event. Queued interactions are resolved instantly — zero interruptions in the agent flow.
Structured system for handling InputRequest and HumanActionRequest. Queue with deduplication (500ms window), automatic expiry of superseded interactions, resolution via terminal with support for multiple choices and default values.
Avatars and documents migrated from local disk to S3. Upload with signed URLs (no public ACLs), presigned URLs for secure download. Backward compatibility with legacy paths.
File attachments for board tasks with S3 storage. Direct upload from the task panel, many-to-many Task ↔ Document association with metadata (size, mime type).
Structured timeline of session events: every action (spawn, prompt, output, commit, error) is tracked with timestamp and metadata. Chronological visualization in the session view.
Feature for linking and navigating similar projects. Suggestions based on workspace and project structure to reuse configurations and prompts across related projects.
Automatic analysis of project files to identify recurring scripts (build, test, lint, deploy). Contextual suggestions based on package.json, Makefile, and CI configurations.
Improved session-workspace linking with automatic fallback and backfill for orphaned sessions. Fixed a sync issue that was incorrectly deleting valid sessions.
Automatic detection of smart actions from session context. Terminal output pattern matching to suggest contextual actions: commit, push, test run, deploy. Quick action panel with one-click execution.
Direct execution of git commands from the session with structured output. Full support for git operations via IPC: arbitrary exec with result parsing, error handling, and configurable timeout.
Automatic generation of optimized prompts via Claude, OpenAI, and Gemini. Project context analysis (modified files, active branch, linked task) to produce targeted prompts. Per-user configurable provider selection.
Task detail side panel with full view: description, assignees, linked sessions, comments, and attachments. Inline editing with auto-save.
Dedicated portal for users with the client role. Filtered view of assigned projects, read-only access to public prompts and tasks. Full permission separation between client and member roles.
Public catalog of prompts shared by the team. Browse by category, full-text search, direct fork into your own workspace. Dedicated page with filters and sorting.
Email notifications for task events (assignment, comment, status change). Threaded comment system on tasks with @user mentions. Per-project notification settings.
Public download page with per-platform assets. Prerelease builds hidden from the public page, visible only to authorized testers. Download tracking with analytics.
Resume interrupted sub-agents with full context. Sub-agent archive with enriched historical logs, resume modal with previous state preview. Project context snapshot for accurate resumption.
Save/restore API for the complete desktop app state. Persistence of window positions, active sessions, open tabs, and provider configuration. Automatic restore after crash or update.
Batch saving of sessions with claude_session_id and associated sub-agents. Efficient synchronization: a single API call to persist the parent session and all child sub-agents.
Full desktop release management: build upload, semantic versioning, multi-platform assets. Download tracking with counts per platform and version. Auto-update URL generated from GitHub Releases.
Anti-DDoS middleware with granular per-endpoint rate limiting. Limits increased 10x for high-traffic APIs (sub-agent, sessions). Brute-force protection on auth and signup.
Automatic 90-day PRO promotion on registration. Tier assignment at user creation with calculated expiry dates. Multi-platform release asset aggregation.
Platform tracking for each desktop installation. Integrated auto-update system with version check, differential download, and silent installation. In-app notification for available updates.
Privacy policy and data processing pages integrated into the app. Legal data seed with versioning. GDPR compliance with user consent management.
Real-time tracking of tokens consumed per session. Baseline estimation (how many would be needed without optimization), savings calculation, and efficiency percentage. Dedicated panel in the session view with detailed breakdown.
At the start of each session, the workspace is analyzed and a project map (structure, entry points, recent files) is injected into the agent context. The agent knows exactly where to go — no blind exploration. Estimated savings: 30K–100K tokens per session.
Sub-agent prompts are automatically enriched with the files actually modified (from git diff) and reduced-scope instructions. The agent reads 5 targeted files instead of exploring 30. Estimated savings: 50K–70K tokens per sub-agent.
Tracking of every file read via hooks with MD5 content hashing. Redundant re-reads (unchanged files) are detected and flagged. Foundation for active interception in upcoming versions.
Robust fallback for tracking: when providers do not expose token counts, the system estimates usage from terminal output volume. Works with any provider and any session.
In the dashboard: total tokens used, tokens saved, average efficiency, and session rankings by savings. Aggregated data from all user sessions with database persistence.
Each session shows which optimizations are active (Smart Context, Cache, Graph, Tool Filtering, Local Execution) with a clear explanation of how each one contributes to savings.
In-app notification system for alerts on prompt sharing, team updates, and session events. Persistent notifications with read/unread state and direct navigation to the relevant resource.
Platform icons (macOS, Windows, Linux) in the hero section are now clickable: a click scrolls to the mailing list signup section with the platform already pre-selected. The newsletter form listens for custom events to receive the selection and automatically populate the platform field.
Every board task becomes executable: it is serialized into a structured prompt with injected context (project, branch, involved files, current sprint) and opens a session with the workspace already mounted. The task → prompt → execution → commit cycle is tracked end-to-end.
Integrated Kanban board with a Sprint model (start/end date, velocity tracking) and Tasks with a state machine: backlog → in_progress → review → done. Many-to-many Task ↔ Session relationship to track which AI sessions worked on which task.
Projects with members, roles (owner, admin, member) and clear permissions. Project model with project_members pivot table and granular RBAC. Cascade delete on member removal. Automatic scoping of all queries by active team and project.
Notes, decisions and documentation live alongside the code. Polymorphic DocNote model linkable to sessions, documents, file paths and tasks. Category-typed documents with full-text search, reusable cross-project.
Open, edit and save files without leaving the session. Built-in editor with syntax highlighting, automatic language detection and file read/write integrated into the desktop process.
Commits generated automatically by analyzing diffs and session activity. Pipeline that analyzes staged files, session context and linked task to generate Conventional Commits with subject, body and metadata.
AI-assisted branch management with AI branch name generation. Full git operations via IPC: stage, unstage, commit, diff, pull, push, stash. Git Explorer with file tree and real-time status.
Automatic connection from .env and read-only data browsing. Auto-detect DB configuration from the project. Multi-database support with schema inspector and tabular queries.
Centralized view with aggregate metrics: active sessions, documents by category, members by role, projects with statistics. Optimized queries with withCount for response times under 200ms.
Consistent work environments across multiple machines. /device-workspaces endpoint with local directory association to remote workspaces by hardware ID. Filesystem path validation on the Electron side, unique index on (path, device_id).
Each project has its own operational space with access to code, files and terminal. Multiple sessions on the same workspace with separate, isolated terminal processes. Advisory locks for conflict management, tabs for fast switching.
Specialized agents (debug, test, security, docs, refactor, performance) coordinated within the same workflow. Each sub-agent has its own dedicated terminal and system prompt. Parent → child hierarchy with spawn, monitoring, output capture and result merging.
Prompts become assets: saved, versioned and tracked based on actual usage. CRUD with diff between versions, fork with derivation chain, dynamic variables {{placeholder}}. Team-level sharing with configurable visibility.
Pro system with approval workflow (request → review → approve/reject). Two-level feature flags: global and per-user. AccessValidator with device status verification and real-time ban/unban.
WebSocket layer for real-time events. Automatic reconnect with exponential backoff. Notifications, activity feed and session state synchronization across clients.
Periodic session state saving every 30 seconds. Automatic recovery after crash or accidental closure. Reusable templates for recurring sessions with pre-configured provider and settings.
Orchestrated and replicable workflows. Configurable hooks on session events (on_start, on_prompt, on_commit, on_end) with shell command execution. Team-level shareable pipelines.
6-digit OTP codes with configurable TTL and rate limiting against brute force. Device tracking with unique hardware ID per session.
Charts for sessions/day, prompts by provider, average session duration, sub-agent distribution by role. Backend aggregation with 15-minute caching.
Unique referral codes per user with invite chain tracking. Assignment at user creation to avoid race conditions.
Claude, Codex, Copilot, Gemini in the same workflow. Core architecture based on isolated sessions, each linked to an AI provider with a dedicated terminal process. Unified interface: spawn, send, receive, terminate.
Full terminal emulator with dynamic resize, Unicode input, ANSI sequences for colors and formatting. Every character written by the AI agent is rendered in real time in the session panel.
Specialized agents for debug, refactor, security and performance. Launched from the main session with parent → child hierarchy. Agent Relay for inter-agent communication with async message passing and acknowledgment.
Saved and reusable prompts in a structured way. Full CRUD with versioning, diff between versions, fork with derivation chain tracking. Dynamic variables {{variable}} and star/bookmark for quick access.
Team management with roles (owner, admin, member), email invites with status tracking (pending → accepted | declined). Team-shared prompts visible to all members.
Persistent timeline of every action: spawn, prompt, output, merge, error. Backend persistence via REST API with time-based filters. Git log-style graph visualization with sub-agent branches.
Service management directly from agents. Production deployment on Docker with Traefik for reverse proxy, SSL termination and load balancing. Automated build with incremental versioning.
Native desktop app with typed IPC architecture for secure main ↔ renderer communication. Dedicated channels for terminal, file system, git and authentication.
REST API with JWT authentication and refresh tokens, resource CRUD, CORS middleware and rate limiting. Primary database with versioned migrations.
Signup with platform tracking (macOS, Windows, Linux) for segmentation. Email validation, deduplication and automated onboarding.
Scegli quali categorie di cookie accettare. I cookie tecnici e funzionali sono sempre attivi.
Per maggiori informazioni, consulta la nostra Privacy e Cookie Policy.
Utilizzati per creare profili relativi all'utente e inviare messaggi promozionali in linea con le preferenze espresse.
Ci permettono di capire come gli utenti navigano il sito per migliorare l'esperienza e i contenuti.
Necessari per il funzionamento del sito. Non possono essere disattivati.
Consentono funzionalità avanzate come la memorizzazione delle preferenze di navigazione.