Skip to content

ROADMAP.md

Latest commit

 

History

History
151 lines (112 loc) · 9.22 KB

ROADMAP.md

File metadata and controls

151 lines (112 loc) · 9.22 KB

ProcessAce Roadmap (Working Draft)

Phase 1 – Skeleton [COMPLETED]

  • Basic Node.js service entrypoint (src/index.js)
  • Health check endpoint
  • Logging utility with structured events
  • Job abstraction (interface + in-memory stub)
  • Simple artifact model with versioning fields

Phase 2 – Ingestion + Jobs [COMPLETED]

  • File upload API (metadata only, stub storage)
  • Enqueue process_evidence jobs
  • Background Worker Implementation
  • Job Lifecycle Events

Phase 3 – LLM Abstraction [COMPLETED]

  • LLM provider interface
  • OpenAI-compatible provider implementation
  • Config via env vars

Phase 4 – Web UI & User Experience [COMPLETED]

  • Served Static Files (Express)
  • Drag & Drop Upload Interface
  • Job Status Polling & Visualization
  • Download links for generated artifacts
  • New: Delete functionality (Job + File + Artifact cleanup)

Phase 5 – BPMN Generation [COMPLETED]

  • Evidence processing worker
  • Prompt Engineering for BPMN 2.0 (JSON graph schema)
  • Deterministic JSON-to-XML compilation (xmlbuilder2)
  • Automatic diagram layout (bpmn-auto-layout)
  • Strict Namespace & Syntax (guaranteed by builder, not LLM)

Phase 6 – Persistence & Reliability [COMPLETED]

  • JSON-based File Store (src/data/*.json)
  • Persistence for Jobs, Evidence, and Artifacts
  • Crash recovery (Jobs survive server restarts)

Phase 7 – Advanced Analysis (SIPOC / RACI) [COMPLETED]

  • SIPOC Matrix Generation (Suppliers, Inputs, Process, Outputs, Customers)
  • RACI Matrix Generation (Responsible, Accountable, Consulted, Informed)
  • Narrative Documentation Generation (Markdown/HTML)
  • UI view for text-based artifacts (Modal & Tables)
  • Custom Artifact Naming & Parallel Generation

Phase 8 – Production Hardening [COMPLETED]

  • Replace JSON FileStore with SQLite (better-sqlite3)
  • Replace In-Memory/JSON Queue with Redis (BullMQ)
  • Structured App-wide Error Handling
  • Docker Containerization (App + Redis via Compose)

Phase 9 – Interactive Visualization & Editing [COMPLETED]

  • Interactive BPMN Viewer: Integrate bpmn-js to view and edit diagrams directly in the browser.
  • Rich Text Editor: Edit Narrative Documentation within the UI (Markdown WYSIWYG).
  • Interactive Tables: Editable SIPOC and RACI grids.
  • Validation Feedback: Real-time validation of edits against BPMN standards (via Modeler).

Phase 10 – Extended LLM Support [COMPLETED]

  • Google GenAI Integration: Support for Gemini models via @google/genai SDK.
  • Anthropic Integration: Support for Claude models via anthropic-sdk.
  • Provider Selection UI: Allow users to switch providers per-project or per-job.
  • Model Selection UI: Allow users to choose specific models (e.g., gpt-5-nano, gemini-3.1-flash-lite-preview).

Phase 11 – Collaboration & Security [COMPLETED]

  • User Authentication: Email/password with JWT (HTTP-only cookies).
  • Multi-User Workspaces: Create workspaces and switch between them.
  • User Data Isolation: Jobs and artifacts scoped per user and workspace (SQLite-backed).
  • Workspace Management UI: Dropdown selector and creation modal in header.
  • Role-Based Access: Viewer / Editor / Admin permissions. First registered user is admin.
  • Admin Dashboard: User management (roles, status) and all-jobs overview with pagination.
  • App Settings: Admin-only LLM configuration UI with encrypted API key storage (AES-256-CBC).
  • User Settings: Profile management (name, password with complexity validation).
  • Process Name Editing: Update process name after job creation.
  • Custom Confirmation Modals: Styled modals replacing native window.confirm().
  • Standardized Header: Consistent header across all pages with user menu.

Phase 12 - Workspace Sharing [COMPLETED]

  • Workspace sharing: Share workspaces with other users.
  • Workspace invitations: Invite users to join workspaces.
  • Member Management: List and remove members from workspaces.
  • Workspace Dashboard: "My Workspaces" and "Shared Workspaces" views with job counts.
  • Workspace Deletion: Owners can delete workspaces and related data.

Phase 13 - Export artifacts [COMPLETED]

  • SIPOC and RACI export: Export SIPOC and RACI matrices to CSV or Excel.
  • BPMN export: Export BPMN diagrams to BPMN or PNG.
  • Narrative export: Export narrative documentation to DOCX or PDF.

Phase 14 - Cleanup codebase [COMPLETED]

  • Refactor codebase: Removed unused debug scripts and temp files, cleaned up .gitignore.
  • Refactor styles: Moved ~50 inline style= attributes from HTML files to style.css using proper CSS classes.
  • Detect reusable code: Extracted shared header into header.js with dynamic injection; replaced native alert() with showAlertModal() from modal-utils.js.
  • Code quality: Added JSDoc comments to backend services and API routes; fixed all ESLint errors.
  • Add documentation: Updated README.md, architecture.md, api_reference.md, user_guide.md, and ROADMAP.md for public release.
  • Package readiness: Updated package.json to v1.0.0, added repository metadata, updated .env.example.
  • Add tests: Added unit tests (settingsService, workspaceService, bpmnBuilder) and integration tests (workspaces, jobs) — 59 tests, 0 failures.

Phase 15 – BPMN Reliability Refactor [COMPLETED]

  • JSON-to-XML Pipeline: LLM outputs structured JSON graph instead of raw XML; deterministic XML compilation via xmlbuilder2.
  • Auto-Layout: bpmn-auto-layout generates <bpmndi:BPMNDiagram> with proper X/Y coordinates.
  • JSON Response Mode: Providers enforce JSON output (OpenAI response_format, Google responseMimeType, Anthropic fence stripping).
  • Zod Schema Validation: src/schemas/bpmnSchema.js — strict runtime validation with .strict() mode, typed enums, JSDoc type inference.
  • Self-Healing Loop: Up to 3 retry attempts; feeds structured Zod error messages back to LLM for correction.
  • Dependencies: Added xmlbuilder2, bpmn-auto-layout, zod.

Phase 16 - Audio Transcription [COMPLETED]

  • Audio Ingestion: Support for uploading audio/video files (mp3, wav, mp4, etc.).
  • Transcription Configuration: Configure a dedicated LLM Provider and Model for transcription (STT) in App Settings involving separate configuration from the main generation model.
  • Transcription Processing: If an audio file is uploaded, automatically transcribe it to text using the configured model.
  • Transcript Review: Review and edit transcripts before generating process artifacts.
  • Transcript Export & Playback: Export transcripts to TXT and play back audio during review.
  • Evidence Pipeline Integration: Use the transcribed text as process evidence to generate artifacts (BPMN, SIPOC, RACI).

Phase 17: Local AI Integration (Ollama)

Phase 17.1: Core Ollama Support & Security

  • UI Enhancements: Add 'Ollama (Local)' to the LLM Provider options (app-settings.html). Implement dynamic UI toggling to hide the API Key field and display a Custom Base URL field when Ollama is selected.
  • Provider Factory Update: Modify src/llm/index.js to securely intercept Ollama selections. Inject a dummy API key to ensure existing OpenAI keys are never accidentally leaked to the local network.
  • SSRF Protection & API Key Relaxation: Update OpenAIProvider to relax the strict API key requirement only for verified local endpoints (e.g., localhost, 127.0.0.1, host.docker.internal). Block external untrusted URLs to prevent Server-Side Request Forgery.
  • Dynamic Model Loading: Verify and test that the existing listModels() implementation successfully fetches available local models via Ollama's OpenAI-compatible /v1/models endpoint.

Phase 17.2: "Plug & Play" Docker Orchestration & Model Manager

  • Infrastructure Update: Integrate the ollama/ollama:latest service into docker-compose.yml. Configure a persistent volume (ollama_data) to ensure downloaded models survive container restarts.
  • Model Pull API: Create a new backend route (/api/settings/llm/pull) that accepts requests to download a curated list of supported models (e.g., Llama 3.2, Qwen 2.5, Mistral).
  • Background Worker Integration: Implement a queue worker (src/workers/modelWorker.js) using the existing Redis/BullMQ architecture to process long-running model downloads via Ollama's native /api/pull endpoint.
  • Model Manager UI: Build a "Local Model Manager" section in the App Settings page. Allow users to select recommended models and initiate downloads directly from the interface.
  • Progress Tracking: Connect the background download worker to the frontend using persisted pull progress and settings-page polling to provide users with a real-time progress bar during model installation.

Phase 17 Notes

  • Ollama is implemented as a generation provider.
  • Local transcription through Ollama is not part of the current runtime and remains deferred until a supported local STT backend is integrated.