TaskForceAI

TaskForceAI

Product updates

TaskForceAI Changelog

Track every release, improvement, and bug fix -- all without leaving your workspace.

Last updated Jul 13, 2026Back to chat

All notable changes across TaskForceAI platforms will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Platform Overview

  • Web Application — TanStack Start multi-agent orchestration platform
  • Desktop Application — Tauri-based macOS/Windows/Linux app
  • Mobile Application — React Native / Expo iOS & Android
  • CLI/TUI — Go-based terminal interface (npm distributed)
  • Server — Pure Go backend with Vercel Functions integration

[Week of 2026-07-13] — Remote Workflows, Cross-Client Parity & Agent Controls

Desktop & Mobile

  • Remote Control: Added account-scoped, explicitly approved mobile-to-desktop control with revocation, presence, Keep Awake, live task state, follow-ups, approvals, files, and change review.
  • Code & Work Shells: Replaced the blocking Code workspace prompt with persistent Code and Work navigation, optional project organization, Code-only repository tools, a split review surface, Open in launchers, and context-window usage.
  • Mobile Remote UI: Added the live Code/Work task timeline, grouped tool activity, inline diffs, shared Changes and Files sheets, camera attachments, pairing by QR or code, project selection, Git review and finish flows, image and web previews, quick actions, and Live Activity routing.
  • Cross-Client Parity: Expanded desktop task navigation, review and terminal workflows; added app-server thread, process, approval, plugin, and compatibility APIs; and brought prompt history, thinking visibility, run controls, markdown, and code timelines to the TUI.
  • Remote Reliability: Persisted command results for durable delivery, hardened relay lifecycle and event polling, recovered pairing and active-host selection, isolated mobile Remote data, and made oversized sync records recoverable.

Agent & Client Workflows

  • Scheduled Tasks: Added native mobile scheduled-task management with safe-area and timestamp fixes.
  • Projects & Finance: Added dedicated project and finance surfaces and multi-repository desktop Code workspaces.
  • Chat Organization: Added pinned chats on web and converted large pasted prompts into attachments.
  • Models & Plugins: Gated premium models by cost tier and added live MCP and skill management to the app-server-backed plugin catalogs.
  • Plan Mode & Skills: Added plan-mode tool gating, session plan persistence, executable skill context, and end-to-end Chat, Code, and Work task-mode propagation.
  • Agent Teams in TUI: Added model-aware team configuration, agent-count persistence, topology labels, and polished model, reasoning-effort, and Agent Teams selectors.

Reliability & Quality

  • Workspace Safety: Required trust for workspace hooks, guarded privileged PTY access, scoped mobile workspace access, and created recovery checkpoints before destructive workspace resets.
  • Streaming & Authentication: Refined realtime turn and tool-progress projection, cleaned up stale streams, and handled expired authentication consistently across runtimes and the TUI.
  • Coverage: Restored the aggregate 100% coverage gate and closed deterministic Mobile Remote, TUI, Go, Rust, and cross-client workflow gaps.

Releases

  • Desktop & Mobile: Advanced the desktop release line to 0.9.12 and the iOS/mobile line to 0.9.4.
  • TUI & app-server: Advanced the TUI to 0.11.23 and released managed app-server 0.11.16.

Key References

[Week of 2026-07-06] — Architecture, Model Controls & Desktop Automation

Architecture & Reliability

  • Clean Architecture: Reorganized packages by architecture ring, moved business policy into core, separated infrastructure and adapters, and expanded cross-language boundary enforcement.
  • Quality Gates: Raised Go test coverage to 100%, strengthened race and static-analysis checks, and made filesystem degradation visible instead of silently masking partial reads.
  • Shared Client Flow: Consolidated task execution, synchronization, and capability wiring across clients while keeping runtime-specific behavior behind adapters.

Models & Reasoning

  • Model Catalog: Updated OpenAI entries to the GPT-5.6 family, added Meta Muse Spark 1.1, migrated xAI references to Grok 4.5, and repaired Gemini, Claude, and Grok routing.
  • Reasoning Effort: Added model-aware reasoning effort controls, moved the web control into the model menu, preserved fallback metadata, and prevented effort settings from carrying across incompatible model changes.

Desktop Auth & Releases

  • Desktop Authentication: Cleared stale desktop auth after submit 401 responses, refreshed desktop auth after device approval, accepted compatibility token responses, and preserved auth state when Keychain persistence fails.
  • Desktop Releases: Repaired the macOS updater path, released desktop 0.5.6, and kept CLI/Desktop version bumps aligned with the tag-driven release flow.
  • Usage Settings: Added shared usage settings surfaces for web and mobile so plan, limit, percentage, and reset labels come from the shared profile view model.

Local Desktop Capabilities

  • Browser & Appshots: Added the embedded desktop browser pane, browser preview diagnostics, review worktrees, appshots, and local desktop prompt routing for browser-open requests.
  • Local Computer Use: Wired local Computer Use observation through the desktop bridge, then gated screen capture to authorized packaged/local desktop origins and Screen Memory permission.
  • Settings Navigation: Added desktop settings search and narrowed desktop local capability controls into searchable sections.

SDK, Mobile & Maintenance

  • SDK Attachments: Aligned TypeScript, Python, Go, and Rust SDK task submissions with the current transient attachment upload contract.
  • Mobile Release Follow-Through: Refreshed mobile TLS pins and advanced the iOS/mobile release line through 0.5.2 and 0.5.3.
  • Mobile Tooling: Refreshed Expo dependencies and tests and advanced NativeWind to the Expo-accepted preview.
  • Maintenance Cleanup: Removed stale scaffolding, tightened production-root dead-code checks, reduced duplicated source/helper/script surfaces, and restored the full repo gates after cleanup fallout.

Key References

[Week of 2026-06-29] — Voice, Privacy & Release Infrastructure

Voice & Mobile App Store

  • Realtime Voice: Added web realtime voice chat, Gateway STT dictation, Gateway TTS playback, browser AudioWorklet capture, authenticated voice setup, and prompt shortcut hardening.
  • Cross-Client Voice: Shared realtime voice runtime logic across web, mobile, desktop, and TUI, including native audio capture/playback, transcript handling, mobile conversation persistence, and TUI voice commands.
  • Mobile Releases: Addressed App Review account and metadata issues, documented App Store availability, prepared mobile 0.3.0, 0.4.0, and 0.5.0 release bumps, and updated marketing iOS links to the App Store.

Privacy, Models & Chat

  • Private Chat: Added Private Chat mode, the no-training disclosure, no-training retention wiring, cross-client UI support, the TUI private-chat command, mobile ordering fixes, and private app-server persistence coverage.
  • Model Selector: Added provider logos to the model selector, bundled them for desktop, hid logos from the compact trigger, and expanded the TUI catalog fallback.
  • Model Routing: Updated the Grok video model route to 1.5 and kept public Sentinel benchmark copy from exposing the backing model name.

Release & Platform Operations

  • CLI & Desktop Releases: Bumped CLI and desktop release versions, fixed Linux CLI ARM64 release dependencies, added Linux ARM64 desktop downloads, sped desktop publishing, fixed macOS updater artifacts, restored desktop Homebrew cask automation, and fixed native self-update behavior.
  • Desktop Updater: Surfaced desktop updater progress and returned no content for current desktop update checks so native update clients handle already-current installs correctly.
  • Mobile Runtime: Updated mobile to Expo 57, fixed store build configuration, hardened mobile auth session persistence, hardened SQLite migrations, and strengthened Apple mobile auth nonce handling.

Security, Reliability & Coverage

  • Security Hardening: Fixed Codex security voice/auth findings, hardened token and MFA handling, blocked disabled users in login flows, removed implicit OpenAI SDK key fallback, and tightened Redis-backed rate limits.
  • Runtime Resilience: Hardened LLM adapters, core approval/tool handling, cache contracts, Redis concurrency, retry classification, sync clocks, private search, PubChem encoding, logging, and Sentry sanitization.
  • Quality Gates: Enforced full TypeScript app/package, Go/Rust service, desktop, TUI, developer, and core coverage; isolated shuffled-test failures; refreshed safe dependencies; and mitigated dependency advisories.

Key References

[Week of 2026-06-22] — Web Reliability & Latency Follow-Through

Web App Startup & Prompt Flow

  • Auth Bootstrap: Cached the authenticated profile hint, bootstrapped the web shell auth/model path earlier, and kept the prompt composer usable during auth hydration.
  • Prompt Preservation: Preserved prompt drafts through hydration and loaded prompt draft capture from the self-hosted script.
  • Sidebar Recovery: Fixed local-first and cloud sidebar history backfills, including partial cache recovery and local pagination preservation.
  • Public Bootstrap: Added app-shell bootstrap snapshot coverage so the public web shell can reuse known-good startup state.
  • Artifact Links: Hid internal artifact metadata from public artifact links and tightened web response validation for public artifact payloads.

Performance & Benchmarks

  • Latency Coverage: Added hosted status, seeded browser, TTFT, P0, and P1 latency benchmarks with baseline comparisons.
  • Runtime Hot Paths: Optimized sync pull, developer proxy, auth refresh, engine active-task indexing, Redis-backed rate-limit checks, and app-server event/SSE parsing.
  • Validation Gates: Hardened benchmark manifests, organized benchmark entrypoints, fixed repo validation fallout, and documented the P1 latency optimization audit.

Security & Maintenance

  • Security Fixes: Fixed sync security findings and added read-media limit regression coverage.
  • Dependency Refresh: Landed the first dependency update wave and related gate cleanup.
  • Admin Usage: Added all-time admin token usage aggregation across the admin API, repository queries, shared SQL, and admin dashboard schemas.
  • QA Harness: Restored the resilience and chaos QA harnesses, added meta-report generation, and covered meta-report parsing.
  • Docs Drift: Updated the documentation drift audit and refreshed OpenAPI generation guidance plus the Go SDK README.
  • Patch Maintenance: Refreshed the safe dependency patch set across Bun, Cargo, Python, and app package manifests.

Key References

[Week of 2026-06-15] — Reliability, Security & Runtime Performance

📱 Mobile & App Review

  • iOS Review Flow: Fixed App Review-facing login and subscription flows, including the "continue without an account" path and mobile billing portal routing.
  • Account Settings: Added mobile authenticator MFA, archived chat management, storage, memory, keyboard, notification, and AI-provider consent settings.
  • Store Builds: Bumped mobile build numbers and hardened React Native runtime pins, worklets, NativeWind env types, and RevenueCat identity handling.

🛡️ Security & Release Hardening

  • Security Findings: Remediated high, medium, low, informational, cloud, mobile, desktop command, MCP, billing, auth, and engine policy findings from the security scan pass.
  • Auth & Rate Limits: Hardened auth session resolution, callback handling, Redis-backed rate limiting, and shared auth/audit paths.
  • Desktop Release: Fixed macOS DMG notarization by preparing the notarytool profile in the release keychain and tightening release workflow checks.

⚡ Engine, Sync & Developer Experience

  • Agent Runtime: Optimized stream chunk assembly, progress tool compaction, tool result packaging, enginecore grep scanning, read-window formatting, glob bounds, and tree rendering.
  • Sync Resolver: Optimized metadata conflict merges and added regression coverage for sync retry and payload edge cases.
  • Quality Gates: Parallelized Go check and test scheduling, added knip dead-code sweep tooling, and expanded full-gate coverage across backend, frontend, SDK, and shared packages.

🌐 Web, Marketing & Docs

  • SEO Crawlability: Fixed marketing changelog, blog, legal, benchmark, and safe-markdown crawlability paths.
  • Deploy Manifests: Added a checked-in Vercel app manifest and schema, plus developer API documentation drift checks.
  • Dependencies: Updated the TypeScript RC package and refreshed the low-risk dependency patch set across Bun, Cargo, and Python lockfiles.

🔗 Key References

[Week of 2026-06-04] — App Polish & Progress Fidelity

🌐 Web & Marketing

  • Benchmarks: Split benchmarks into a dedicated section and refreshed the public model comparison copy.
  • Changelog: Updated the changelog page layout so long release notes render fully instead of clipping inside the panel.
  • Model Labels: Ensured Sentinel is presented as TaskForceAI's public model across progress and model-selection UI.

🤖 Agent Experience

  • Progress Bubble: Improved sparse progress smoothing so long-running work feels active without claiming completion early.
  • Tool Usage: Matched live tool events more reliably to agent rows, including custom model labels and single-agent runs.

[Week of 2026-01-19] — Speed & Polish

⚡ Core Engine

  • Quick Mode: Introduced a new mode to bypass multi-agent orchestration for instant, direct model responses.
  • Performance: Optimized server response times for non-complex queries.

📚 Documentation

  • User Guide: Updated USER_FEATURES.md with Quick Mode instructions.

[Week of 2025-12-30] — New Year, New Stack

🏗️ Architecture & Ops

  • Web Migration: Migrated the web application to TanStack Start
  • Server Port: Completed transition to a pure Go server
  • Marketing Split: Extracted marketing site into apps/marketing
  • Vercel Functions: Implemented Go runtime support

🌐 Web & Marketing

  • Added OG image generation using @vercel/og
  • Implemented full-page navigation
  • Decoupled authentication logic from framework-specific dependencies

[Week of 2025-12-23] — The Go Rewrite & Clean Architecture

⚡ Core Engine

  • Go Port: Ported orchestration engine and shared packages (100% coverage)
  • SOLID Refactor: Clean Architecture refactor across backend
  • Dependency Injection: DIP for LLM clients, caching, repositories

🧪 Testing

  • Achieved 97% test coverage
  • Standardized test suites across packages

[Week of 2025-12-16] — UI/UX Polish & E2E Stability

🎨 Frontend

  • AI-powered background removal for logos
  • Improved voice control resilience with stop capability
  • Refreshed landing page and blog

🤖 Automation

  • Stabilized Detox / Playwright E2E tests
  • Met 80%+ frontend coverage targets

[Week of 2025-12-09] — Modularization

🧩 Refactoring

  • Broke monoliths into focused components (<300 LOC)
  • Refactored agent orchestration & search tooling
  • Reorganized build & maintenance scripts

[Week of 2025-12-02] — Toolchain & Performance

🛠️ Developer Experience

  • Migrated from ESLint → Oxlint
  • Prepared for TS7 features
  • Repo-wide clean code & logging pass

[Week of 2025-11-25] — Type Safety & Models

🛡️ Safety

  • Enforced strict DB structs in Go/Desktop
  • Python SDK modularized with Pydantic
  • Shared validation packages added

[Week of 2025-11-18] — Testing & Logging

🧪 Quality Assurance

  • Migrated tests to Bun
  • Structured JSON logging everywhere
  • Cleaned agent/environment docs

[Week of 2025-11-11] — Infrastructure & Gateway

☁️ Cloud & Data

  • Migrated to Vercel AI Gateway
  • Integrated Upstash Redis & tRPC v11
  • Mobile UI aligned for NativeWind v5

[Week of 2025-11-04] — Cross-Platform Expansion

🖥️ Desktop & CLI

  • Windows/Linux desktop builds
  • Cross-platform CLI binaries
  • Installer distribution via Vercel Blob
  • CLI v0.6.2 released

📱 Mobile

  • Secure storage & Google OAuth
  • Offline-first sync improvements
  • CI minutes reduced ~75%

[0.1.1] — 2025-11-04

🖥️ Desktop Application (Tauri)

  • Bundled the standalone web server with embedded SQLite
  • Fixed Dexie initialization & pnpm symlink issues

[1.0.0] — 2025-11-03

🚀 Initial Major Release

  • Web: Multi-agent system, TanStack Start, Stripe, Redis, Prisma
  • Desktop: Native macOS app with Sparkle updates
  • Mobile: React Native (Expo SDK 54)
  • CLI: Go TUI (Bubble Tea)
  • SDKs: TypeScript & Python
  • API: Public REST API with tiered pricing

Release Process & Links