Roadmap

R3 is now live. See what shipped, what's coming next, and what's in the backlog.

Production · Stable

Deterministic, self-guarding execution. Every workflow runs as a tracked, resumable plan, guardrails and advisories fire at the moment of action across every IDE, and a leaner bootstrap keeps each session fast.

Deterministic Execution
  • Every workflow runs as a tracked, resumable plan — create, advance, and audit each step
  • Sequential phase gates keep the agent on process: preparation, guardrails, HITL, and risk checks are explicit steps
  • Plan and act modes supported directly, with atomic writes and resume-after-failure
  • Graceful fallback to built-in task tracking when the operation manager is unavailable
Cross-IDE Guardrails & Advisories
  • A hooks runtime enforces guardrails and advisories at the moment of action
  • Self-reviewable dangerous-actions gate on destructive shell, file, and MCP operations — safe across worktrees and parallel sessions
  • Advisory nudges for markdown placement, syntax/type/lint/format steps, and stray module files
  • One input adapter normalizes Claude Code, Cursor, Copilot, Codex, and Windsurf — author a hook once, run it everywhere
Leaner Bootstrap
  • Per-session bootstrap roughly half its former size through skill extraction and de-duplication
  • Progressive disclosure — thin skills load heavy assets only on demand
  • Self-initializing preparation: agents load context, select the workflow, and commit to phases before acting
Automated QA & Test Generation
  • Expanded automated-QA flow family — dedicated UI and API AQA flows plus test-case generation
  • Structured phases from data collection and requirements clarification through selector implementation, execution, and test correction
Code Graph & Deployment
  • GitNexus code-graph integration for architecture-aware navigation and automatic reindexing
  • Public OSS Rosetta MCP and RAGFlow deployment for open-source and demo environments
  • GitHub authentication for MCP access
Previous Release · Backports Only

Enterprise-grade release with flexible deployment, built-in security, an expanded workflow library, and intelligent agent orchestration. Backwards-compatible with R1.

Deploy Anywhere
  • Multiple connection modes — cloud, local, plugin, and fully offline
  • Scales horizontally for teams of any size
  • Air-gap capable — runs inside your perimeter with no external dependencies
  • Container-ready with Docker Compose and Kubernetes support
Enterprise Security
  • Industry-standard authentication with support for major identity providers
  • Tokens encrypted at rest; policy-based access control per team and project
  • Human-in-the-loop approval gates at every critical decision point
  • Sensitive data protection, risk assessment, and scope enforcement built in
  • Rosetta is designed so that no source code leaves your organization
Workflows & Skills
  • 14+ structured workflows covering the full SDLC — from initialization to validation
  • 30 skills spanning coding, testing, planning, research, security, and documentation
  • Every request auto-classified and routed to the right workflow
  • Context loads progressively — agents get only what they need, when they need it
  • Five-phase execution: Prepare, Research, Plan, Act, Validate
Intelligent Agents
  • Specialized agents for discovery, planning, implementation, review, and validation
  • Persistent execution plans — resume where you left off, even across sessions
  • Parallel delegation with scoped context for large workspaces
  • Automatic bootstrapping — project context loads before the agent acts
Knowledge Architecture
  • Three-layer instructions: core standards + organization rules + project-specific context
  • Instructions versioned and published as code — rollback anytime
  • Automatic workspace documentation generated during onboarding
  • Incremental publishing — only changed instructions are updated
Developer Experience
  • One-file bootstrap — single rule file gets any IDE connected
  • Works across 9 IDEs: Cursor, Claude Code, VS Code, Windsurf, JetBrains, Codex, Antigravity, OpenCode
  • CLI with dry-run, offline mode, and fast incremental updates
  • Works with any MCP-compatible tool — present and future
Out of Support

Delivered the core Rosetta experience — centralized knowledge delivery, automated repo initialization, and structured workflows across all major AI coding IDEs.

IDE Integration
  • One-command setup across Cursor, Claude Code, VS Code, Windsurf, JetBrains, and any MCP-compatible tool
  • Automatic knowledge retrieval and context injection into every AI session
Repository Onboarding
  • Automated analysis of tech stack, dependencies, and architecture
  • Project documentation generated in minutes, not days
  • Agent rules and workflows configured per project
Knowledge & Workflows
  • Centralized knowledge base with rules, guardrails, and best practices
  • CLI for publishing and managing instructions as code
  • Prepare → Research → Plan → Act → Validate execution model with task classification

We're focused on R3. The backlog captures directions we're exploring — not commitments. Priorities shift based on community feedback and production learnings.

Quality
  • Automated quality scoring for instructions and workflows
  • Before-and-after evaluation when instructions change
Scale
  • Multi-agent parallel execution for large tasks
  • Cross-project awareness — catch breaking changes before they ship
Community
  • Open contribution model for workflows and skills
  • Self-hosting guides and expanded integrations