Changelog

v0.1.0: Initial Release

1 April 2026

Pipeline-First Development

  • Goal to spec to task to agent workflow
  • Functional spec, test spec and security spec stages
  • Test and security specs run in parallel after spec approval
  • Human-in-the-loop review at every gate
  • Automatic task decomposition from natural language goals
  • Pause and resume for active goals

Multi-Agent Orchestration

  • Claude, Codex and Gemini CLI support
  • Per-stage agent assignment via pipeline profiles
  • Automatic CLI detection across 40+ installation methods
  • Adaptive concurrency with per-adapter rate limiting
  • Mission Mode (Cmd+M) for instant, single-agent tasks (Standard, Research, Audit and Plan modes)
  • Auto Model: intelligent adapter and model routing based on task intent
  • Plans: step-by-step mission outputs promotable to full Goals

Remote Execution

  • SSH-based remote agent execution on any accessible host
  • Remote project management with the same worktree isolation model as local projects
  • Real-time transcript streaming from remote agents to the local UI
  • Uses existing SSH keys; no private keys stored by Dispatch

Sandbox Isolation

  • All agent writes happen in a git worktree mirror of your project
  • The real project is never directly modified during agent runs
  • Discard sandbox on rejection with no side effects
  • Worktree symlink optimization for large monorepos

Paper Import

  • Implement research papers directly: paste a URL or upload a PDF
  • Dispatch extracts the methodology and generates a pre-populated spec

Agent-Assisted Review

  • Full-viewport review workspace with Summary, Checklist, Diff and Conversation tabs
  • Spec verification checklist gates approval on block-tier reviews
  • Risk classification of files (high, medium, low)
  • Anti-rubber-stamp measures for block-tier approvals

Secure by Design

  • OWASP Top 10 mapped security specs
  • Post-build security audit (holistic, cross-task)
  • Post-build quality audit (performance, scalability and reliability)
  • Destructive command detection in agent transcripts
  • Dual-agent security review (opt-in)

Context Management

  • Compaction detection (behavioral drift + explicit signals)
  • Structured 9-section compaction summaries
  • Post-compaction file recovery with token budgets
  • Microcompaction for incremental summarization
  • Session memory extraction for cross-run learning
  • Decision tracking and replay timeline in the Vault

Developer Experience

  • Dashboard with personalized greetings and animated micro-illustrations
  • 3D Command Center with fullscreen HUD and interactive cubicles
  • Decision Query (Lore protocol) for file-level decision history
  • Real-time agent monitoring with reasoning preview
  • Mid-task steering (corrections while agents work)
  • Token efficiency monitoring with cache hit ratios and red flag alerts
  • In-app preview for completed goals (spawns dev server)
  • Native macOS menus and keyboard shortcuts
  • CLI with 17 commands (dispatch bundled with the desktop app)
  • Retry logic with exponential backoff on transient API failures
  • Performance dashboard with adapter metrics and correction management
  • Multi-mission tabs with concurrent execution

Remote Access

  • HTTPS browser access with auto-generated self-signed TLS certificate
  • Access Dispatch from any device on your network

Quality Assurance

  • 2,100+ automated tests (server, adapter, CLI unit tests plus Playwright e2e)
  • Diminishing returns detection (auto-stop looping agents)
  • Per-task token budget enforcement
  • Auto-approval rules for low-risk changes
  • Worktree symlink optimization for large projects

System Requirements

  • macOS 12+ (Apple Silicon or Intel) or Linux (x86_64 or ARM64)
  • At least one agent CLI installed: Claude Code, Codex or Gemini CLI
  • Bun runtime (bundled in the app)

Full release notes and assets on GitHub Releases.