Skip to content

ROADMAP.md

Latest commit

 

History

History
528 lines (454 loc) · 24.5 KB

ROADMAP.md

File metadata and controls

528 lines (454 loc) · 24.5 KB

Archon - Project Roadmap

Living document tracking all phases, milestones, and deliverables. Incorporates all 20 strategic improvements and 10 build-correctness strategies.

Phase Overview

Phase 0 --> Phase 1 --> Phase 2 --> Phase 3 --> Phase 4 --> Phase 5 --> Phase 6 --> Phase 7
Vertical    Core        Ops &       Security    Integ &     Deploy &    Advanced    Validation
Slice       Platform    Cost        & Gov       Data        UX          Features    & Polish
(Orch)      (01-06)     (07-09,23)  (10-12,     (13,14,19)  (15-17,22)  (24,25)     (Master
                                     18,20,21)                                       Validator)

Phase 0: Foundation & Vertical Slice (Orchestrator - MANDATORY FIRST)

Goal: Prove the entire architecture works end-to-end before any agent scales out. Implements Build-Correctness Strategies 1-6, 8-9.

Milestone 0.1 - Contract-First Artifacts (Strategy 1)

  • Generate contracts/openapi.yaml - complete OpenAPI 3.1 spec for all planned endpoints
  • Generate contracts/events.yaml - WebSocket event schemas (JSON Schema)
  • Generate contracts/shared-types.ts - TypeScript types auto-generated from OpenAPI
  • Lock contracts - agents implement against these, no deviations allowed

Milestone 0.2 - Architecture Decision Records (Strategy 2)

  • docs/ADR/001-api-response-format.md
  • docs/ADR/002-error-handling-pattern.md
  • docs/ADR/003-database-naming.md
  • docs/ADR/004-task-queue-pattern.md
  • docs/ADR/005-auth-middleware.md
  • docs/ADR/006-logging-format.md
  • docs/ADR/007-testing-patterns.md
  • docs/ADR/008-websocket-protocol.md
  • docs/ADR/009-env-config-pattern.md
  • docs/ADR/010-import-structure.md

Milestone 0.3 - Interface Stubs & Golden Path (Strategies 3, 7)

  • backend/app/interfaces/ - Abstract base classes for all cross-agent boundaries
  • backend/app/interfaces/models/ - Shared Pydantic models (Agent, Execution, User, etc.)
  • docs/golden-path/router-example.py
  • docs/golden-path/service-example.py
  • docs/golden-path/test-example.py
  • docs/golden-path/model-example.py

Milestone 0.4 - Integration Test Contracts (Strategy 4)

  • tests/integration/ - Pre-written failing tests at every agent boundary
  • Backend to Frontend contract tests (API response shapes)
  • Router to LLM Provider contract tests (model interface)
  • Connector to Backend contract tests (data source interface)
  • Agent to Execution Engine contract tests (state machine interface)

Milestone 0.5 - Build Verification Pipeline (Strategies 5, 8, 10)

  • Makefile with make verify (syntax, imports, types, lint, tests, contract diff)
  • make test-slice - heartbeat test for the vertical slice
  • CI/CD pipeline definition (GitHub Actions) with regression guardian
  • docs/SELF_VERIFICATION_CHECKLIST.md - mandatory agent pre-completion gate

Milestone 0.6 - Vertical Slice (Strategy 9)

  • End-to-end: "Create and Execute a Simple 2-Node Agent"
    • Frontend: drag 2 nodes, connect, click Run
    • API: POST /agents, POST /execute
    • Auth: JWT validation via Keycloak
    • LangGraph: execute 2-node state machine
    • WebSocket: stream execution output
    • Database: persist agent definition + execution result
    • Audit: log events with correlation IDs
    • Cost: record token usage
    • Docker: docker-compose up runs everything
  • Vertical slice stays green at ALL times - make test-slice is the heartbeat

Phase 1: Core Platform (Agents 01-06)

Goal: Functional backend + UI builder + NL wizard + templates + sandbox + versioning

Milestone 1.1 - Core Backend (Agent-01)

  • FastAPI project structure with modular routers
  • SQLModel schemas for all core entities (agents, versions, executions, users)
  • Alembic migration setup with initial schema
  • LangGraph integration - agent state machine execution engine
  • WebSocket server for real-time builder updates and streaming
  • OpenTelemetry instrumentation (traces, metrics, logs)
  • Docker Compose for local development (API + Postgres + Redis + MinIO)
  • Health check, readiness, and liveness endpoints
  • API key management (creation, rotation, revocation)
  • Basic RBAC middleware
  • [Improvement 7] Data retention & lifecycle policy engine
  • [Improvement 7] Right-to-erasure workflows (GDPR)
  • [Improvement 7] Data lineage tracking

Milestone 1.2 - No-Code Builder UI (Agent-02)

  • React 19 + Vite + TypeScript project setup
  • React Flow canvas with custom node types
  • Node palette (200+ nodes organized by category)
  • Property panel with context-aware configuration
  • Live preview pane with streaming agent output
  • Version timeline component with diff view
  • Export to JSON and Python
  • Responsive layout with dark/light mode
  • Keyboard shortcuts and accessibility
  • Integration with backend WebSocket API
  • [Improvement 18] Visual Agent Debugger - breakpoints, step-through, state inspect
  • [Improvement 18] Time-travel debugging (go backwards through execution)
  • [Improvement 18] State modification mid-execution and resume

Milestone 1.3 - Natural Language to Agent Wizard (Agent-03)

  • Multi-step LangGraph workflow: Describe > Plan > Build > Validate
  • Planner LLM generates agent specification and flow diagram
  • Coder LLM generates LangGraph JSON + Python from spec
  • Validator LLM checks generated agent for security issues
  • User feedback loop with iterative refinement
  • Auto-deploy to sandbox after validation passes
  • Natural language to node suggestion (inline in canvas)

Milestone 1.4 - Template Library (Agent-04)

  • Template data model (metadata, categories, tags, popularity)
  • Template CRUD API endpoints
  • Template browser UI with search, filter, preview
  • One-click deploy from template
  • Fork and customize templates
  • Community contribution workflow (submit, review, publish)
  • GitHub sync for template repository
  • 50 initial templates across categories

Milestone 1.5 - Sandbox & Testing (Agent-05)

  • Isolated Docker execution environment per test run
  • Kubernetes namespace isolation for sandbox workloads
  • Arena Mode - parallel agent execution with metrics comparison
  • Test suite generation from agent definition
  • Automated regression testing on version changes
  • Resource limits and timeout enforcement
  • Test results dashboard with comparison charts
  • [Improvement 17] Standardized agent benchmarking suite
  • [Improvement 17] Public leaderboard for community agents
  • [Improvement 17] Enterprise private benchmarks

Milestone 1.6 - Version Control (Agent-06)

  • Git-like versioning for agent definitions
  • Immutable version snapshots (code, config, dependencies)
  • Visual diff between versions
  • One-click rollback to any previous version
  • Branch and merge support for agent development
  • Deployment promotion (dev > staging > production)
  • Audit trail for all version changes

Phase 2: Operations & Cost (Agents 07, 08, 09, 23)

Goal: Intelligent routing, lifecycle management, cost tracking, multi-tenancy

Milestone 2.1 - Intelligent Router (Agent-07)

  • Dynamic model selection engine (cost x latency x capability x sensitivity)
  • Real-time latency and availability monitoring per provider
  • Fallback chains with automatic failover
  • Per-department and per-user routing rules
  • A/B testing support for routing strategies
  • Router configuration UI
  • < 200ms p95 routing decision latency
  • LiteLLM integration for unified model access
  • [Improvement 13] Explainable routing - human-readable decision with every route
  • [Improvement 13] Decision audit trail and replay/simulation
  • [Improvement 13] "Why was this model chosen?" UI

Milestone 2.2 - Lifecycle Manager (Agent-08)

  • Model registry with approval workflows
  • Canary deployments for agent updates
  • Blue-green deployment support
  • Automated rollback on error rate thresholds
  • Agent health monitoring and auto-restart
  • Scheduled agent execution (cron-like)
  • Agent dependency graph visualization
  • [Improvement 14] Real-time streaming ops dashboard
  • [Improvement 14] ML-based anomaly detection
  • [Improvement 14] Predictive observability (budget forecasting, error trending)
  • [Improvement 14] Auto-mitigation (throttle, failover, pause on anomaly)

Milestone 2.3 - Cost Engine (Agent-09)

  • Universal token ledger tracking all LLM usage
  • Real-time cost dashboard per agent, user, department
  • Departmental budgets with alerts and hard limits
  • Cost forecasting based on historical trends
  • Chargeback reports (PDF + API)
  • Cost optimization recommendations
  • OpenLLMetry integration for standardized tracking
  • Multi-currency support

Milestone 2.4 - Multi-Tenant Platform & Billing (Agent-23) NEW

  • [Improvement 12] Tenant isolation (database RLS, compute namespaces, network policies)
  • [Improvement 12] Self-service onboarding: signup > verify > choose tier > provision
  • [Improvement 12] Tier definitions: Free > Individual > Team > Enterprise
  • [Improvement 12] Usage metering (executions, tokens, storage, API calls)
  • [Improvement 12] Stripe billing integration (subscriptions + usage-based)
  • [Improvement 12] Internal chargeback mode (no external billing dependency)
  • [Improvement 12] Quota enforcement with configurable hard/soft limits
  • [Improvement 12] Tenant admin console

Phase 3: Security & Governance (Agents 10, 11, 12, 18, 20, 21)

Goal: Adversarial testing, DLP, compliance, shadow AI discovery, MCP security, proxy

Milestone 3.1 - Red-Teaming Engine (Agent-10)

  • Garak integration for automated adversarial testing
  • Custom attack vector library (prompt injection, jailbreak, data exfil)
  • Scheduled red-team runs per agent
  • Vulnerability scoring and trending
  • Remediation recommendations
  • Integration with CI/CD pipeline
  • Report generation (executive + technical)

Milestone 3.2 - DLP & Guardrails (Agent-11)

  • Multi-layer DLP pipeline (regex > NER > semantic > policy)
  • Presidio integration for PII/PHI detection
  • Custom entity recognizers (API keys, internal IDs, etc.)
  • NeMo Guardrails for content safety
  • Guardrails AI for structural output validation
  • Configurable policies per agent/department
  • Real-time DLP dashboard with alerts
  • Hallucination detection (cross-reference + confidence scoring)
  • Toxicity and bias scoring
  • [Improvement 19] Natural Language Policy Engine - define policies in plain English
  • [Improvement 19] Auto-generate OPA + DLP rules from NL description
  • [Improvement 19] Policy impact simulation

Milestone 3.3 - Governance Dashboard (Agent-12)

  • Neo4j graph for agent/model/data lineage
  • Compliance dashboard (SOC2, GDPR, HIPAA status)
  • Agent registry with ownership and approval status
  • Data flow visualization
  • Policy management UI (OPA policies)
  • Audit log viewer with advanced search and filters
  • Export compliance reports (PDF, CSV)
  • Role-based access to governance features
  • [Improvement 5] Auto-generate reports for EU AI Act, NIST AI RMF, ISO 42001
  • [Improvement 5] Pluggable regulation templates (community-contributed)
  • [Improvement 8] AI Inventory Management with custom risk tiers
  • [Improvement 8] Dynamic risk scoring based on usage patterns
  • [Improvement 9] Governance committees with approval workflows
  • [Improvement 9] Time-bounded approvals with auto-escalation
  • [Improvement 15] Agent dependency graph with blast radius analysis
  • [Improvement 15] Change impact simulation

Milestone 3.4 - SentinelScan Engine (Agent-18) NEW

  • [Improvement 1] Shadow AI discovery across the organization
  • [Improvement 1] SSO audit log scanner, network traffic analyzer, API gateway inspector
  • [Improvement 1] Unified AI asset inventory (cross-platform, not just Archon agents)
  • [Improvement 1] Risk classification with configurable tiers
  • [Improvement 1] Real-time posture score (0-100) with trend tracking
  • [Improvement 1] Executive summary export (PDF)
  • [Improvement 1] Pluggable discovery module SDK

Milestone 3.5 - MCP Security & Governance (Agent-20) NEW

  • [Improvement 3] Ephemeral sandboxed containers per MCP request (zero data retention)
  • [Improvement 3] Tool-level authorization and access control
  • [Improvement 3] Tool definition change detection with side-by-side diff
  • [Improvement 3] Response validation + indirect prompt injection detection
  • [Improvement 3] Community-driven MCP vulnerability database
  • [Improvement 3] Version pinning and rollback for MCP tools

Milestone 3.6 - Security Proxy Gateway (Agent-21) NEW

  • [Improvement 4] Standalone reverse-proxy for ANY AI endpoint
  • [Improvement 4] DLP pipeline: PII detection/redaction on requests and responses
  • [Improvement 4] Content classification and policy enforcement (OPA)
  • [Improvement 4] Token counting and cost attribution across all proxied services
  • [Improvement 4] Full audit logging with SIEM export
  • [Improvement 4] Streaming SSE passthrough for chat completions
  • [Improvement 4] Standalone Docker deployment with YAML config
  • [Improvement 4] < 50ms added latency (p95)

Phase 4: Integrations & Data (Agents 13, 14, 19)

Goal: 50+ connectors, document processing, A2A protocol, RAG pipeline

Milestone 4.1 - Connector Hub (Agent-13)

  • Connector framework with plugin architecture
  • OAuth2 credential management for connectors
  • [Improvement 10] Three-tier integration architecture:
    • Data Source Connectors (read-only: SharePoint, Confluence, Drive, etc.)
    • MCP Servers (bidirectional tool access)
    • Tool Integrations (action-oriented: Jira actions, email sending, etc.)
    • Event Sources (webhooks, Kafka, DB change streams - 4th category)
  • Initial connectors: M365, Google Workspace, Salesforce, Slack, Jira/Confluence, GitHub/GitLab, databases, cloud storage, ServiceNow, generic REST/GraphQL/gRPC
  • [Improvement 6] Permissions-aware data access (inherit source app permissions)
  • [Improvement 6] Transparent permissions cache with real-time sync
  • [Improvement 6] Audit logging of every permission check
  • Connector health monitoring, rate limiting, retry logic
  • Custom connector SDK (Python)
  • Connector marketplace UI

Milestone 4.2 - DocForge Processor (Agent-14)

  • Unstructured.io integration for multi-format parsing
  • Parallel processing pipeline (chunks > embeddings > store)
  • Automatic document tagging and classification
  • OCR for images and scanned documents
  • Table extraction and structured data output
  • Document change detection and re-indexing
  • Processing status dashboard
  • Support for: PDF, DOCX, PPTX, XLSX, HTML, MD, TXT, Images

Milestone 4.3 - A2A Protocol Support (Agent-19) NEW

  • [Improvement 2] Agent Card discovery from well-known URIs
  • [Improvement 2] Import external A2A agents as drag-and-drop canvas nodes
  • [Improvement 2] Publish Archon agents as A2A-compatible services
  • [Improvement 2] Bi-directional task delegation with streaming
  • [Improvement 2] mTLS + OAuth2 authentication for cross-platform calls
  • [Improvement 2] Full audit trail of all A2A interactions

Phase 5: Deployment & UX (Agents 15, 16, 17, 22)

Goal: Mobile SDK, interactive components, marketplace, production deployment

Milestone 5.1 - Live Components (Agent-15)

  • WebSocket protocol for embedded React components in chat
  • Component library (forms, charts, tables, buttons, cards)
  • Dynamic component rendering based on agent output
  • State management between chat and embedded components
  • Component SDK for custom interactive elements
  • Theme support (inherits from host application)

Milestone 5.2 - Mobile SDK (Agent-16)

  • Flutter SDK package for agent interaction
  • iOS native chat application
  • Android native chat application
  • Push notifications for agent updates
  • Offline mode with sync
  • Biometric authentication
  • Voice input/output support
  • Responsive agent cards and rich media

Milestone 5.3 - Production Deployment (Agent-17)

  • Helm chart with configurable values for all services
  • Terraform modules for AWS, Azure, GCP
  • Air-gapped deployment bundle (all images + charts)
  • ArgoCD application definitions
  • Kyverno policies for security enforcement
  • Cert-Manager configuration for automatic TLS
  • Monitoring stack (Prometheus + Grafana + AlertManager)
  • Logging stack (OpenSearch + Fluent Bit)
  • Backup and restore procedures
  • Disaster recovery runbook
  • Performance tuning guide
  • Security hardening checklist

Milestone 5.4 - Open Marketplace & Creator Program (Agent-22) NEW

  • [Improvement 11] Listing system: agents, templates, connectors, guardrail policies
  • [Improvement 11] Search with full-text + faceted filtering
  • [Improvement 11] Creator registration, profiles, verified badges
  • [Improvement 11] Submission > automated review > manual review > publish pipeline
  • [Improvement 11] One-click install from marketplace to workspace
  • [Improvement 11] Stars, reviews, forks, usage analytics
  • [Improvement 11] Self-hostable private marketplace for enterprises
  • [Improvement 11] Optional Stripe Connect for premium listings

Phase 6: Advanced Features (Agents 24, 25)

Goal: Cross-org collaboration and edge/offline deployment

Milestone 6.1 - Federated Agent Mesh (Agent-24) NEW

  • [Improvement 16] Cross-org agent communication with mutual TLS
  • [Improvement 16] Mesh discovery and trust establishment
  • [Improvement 16] Data isolation with policy-gated sharing
  • [Improvement 16] End-to-end encryption for all mesh messages
  • [Improvement 16] Audit trails on both sides of cross-org interactions
  • [Improvement 16] Kill switch for immediate trust revocation
  • [Improvement 16] Federation agreement management

Milestone 6.2 - Edge Runtime (Agent-25) NEW

  • [Improvement 20] Lightweight runtime (< 500MB) for edge devices
  • [Improvement 20] Local model inference (Ollama, vLLM, ONNX Runtime)
  • [Improvement 20] Local vector store for offline RAG
  • [Improvement 20] Bi-directional sync with conflict resolution
  • [Improvement 20] Embedded OPA + local DLP scanning
  • [Improvement 20] Device authentication and tamper detection
  • [Improvement 20] Fleet management console on central platform
  • [Improvement 20] Remote wipe capability

Phase 7: Validation & Polish (Master Validator)

Goal: 95%+ feature coverage, E2E testing, documentation, launch readiness

Milestone 7.1 - E2E Testing (Master Validator)

  • 50 enterprise scenario test suite
  • Performance benchmarking under load (10k concurrent agents)
  • Security penetration testing
  • Compliance audit simulation
  • Cross-browser testing (Chrome, Firefox, Safari, Edge)
  • Mobile app testing (iOS + Android)
  • API contract testing
  • Chaos engineering tests (pod failures, network partitions)

Milestone 7.2 - Documentation

  • Getting Started guide
  • API Reference (auto-generated from OpenAPI)
  • Architecture overview
  • Security whitepaper
  • Deployment guides (Docker, K8s, Air-gapped, Cloud, Edge)
  • Connector development guide
  • Agent development tutorials
  • Video walkthroughs
  • FAQ and troubleshooting

Milestone 7.3 - Launch Readiness

  • Open-source community setup (CONTRIBUTING.md, CODE_OF_CONDUCT.md)
  • GitHub repository configuration (issues, discussions, CI/CD)
  • Demo environment (publicly accessible)
  • Marketing site / landing page
  • Blog post / announcement
  • ProductHunt launch plan

Build-Correctness Strategy Traceability

Every strategy from docs/BUILD_CORRECTNESS.md is mapped to milestones above.

Strategy Milestone Status
1. Contract-First API Specs Phase 0, M0.1 Planned
2. Binding ADRs Phase 0, M0.2 Planned
3. Interface Stubs Phase 0, M0.3 Planned
4. Integration Test Contracts Phase 0, M0.4 Planned
5. Build Verification Pipeline Phase 0, M0.5 Planned
6. Dependency-Ordered Execution Enforced by Orchestrator (MAIN_PROMPT.md) Planned
7. Golden Path Examples Phase 0, M0.3 Planned
8. Self-Verification Checklist Phase 0, M0.5 Planned
9. Vertical Slice First Phase 0, M0.6 Planned
10. Regression Guardian Phase 0, M0.5 (CI/CD definition) Planned

Improvement Traceability

Every improvement from docs/GAP_ANALYSIS.md is mapped to milestones above.

# Improvement Phase Milestone
1 SentinelScan (Shadow AI Discovery) 3 M3.4
2 A2A Protocol Support 4 M4.3
3 MCP Security & Governance 3 M3.5
4 Security Proxy Gateway 3 M3.6
5 Compliance Report Generator 3 M3.3
6 Permissions-Aware Data Pipeline 4 M4.1
7 Data Retention & Lifecycle 1 M1.1
8 AI Inventory + Risk Classification 3 M3.3
9 Governance Committees 3 M3.3
10 Three-Tier Integration Architecture 4 M4.1
11 Open Marketplace & Creator Program 5 M5.4
12 Multi-Tenant & Billing 2 M2.4
13 Explainable Routing 2 M2.1
14 Observability & Anomaly Detection 2 M2.2
15 Dependency Blast Radius 3 M3.3
16 Federated Agent Mesh 6 M6.1
17 Agent Benchmarking Suite 1 M1.5
18 Visual Agent Debugger 1 M1.2
19 Natural Language Policy Engine 3 M3.2
20 Edge/Offline Runtime 6 M6.2

Dependencies Graph (25-Agent Roster)

                    Phase 0: Orchestrator
                    Contracts, ADRs, Stubs,
                    Golden Path, Vertical Slice
                              |
  ============================|============================
  |                           |                           |
Agent-01 (Core Backend) ------+---------------------------+
  |                                                       |
  +--- Agent-02 (UI + Debugger)                           |
  |      +--- Agent-04 (Templates)                        |
  |      +--- Agent-12 (Governance)                       |
  |      +--- Agent-15 (Live Components)                  |
  |      +--- Agent-22 (Marketplace)                      |
  |                                                       |
  +--- Agent-03 (NL Wizard)                               |
  |                                                       |
  +--- Agent-05 (Sandbox + Arena)                         |
  |      +--- Agent-10 (Red-Team)                         |
  |                                                       |
  +--- Agent-06 (Versioning)                              |
  |                                                       |
  +--- Agent-07 (Router + XAI)                            |
  |      +--- Agent-08 (Lifecycle + Anomaly)              |
  |      +--- Agent-09 (Cost Engine)                      |
  |      +--- Agent-11 (DLP + NL Policy)                  |
  |      +--- Agent-21 (Security Proxy)                   |
  |                                                       |
  +--- Agent-13 (Connectors 3-Tier) --- Agent-14 (DocForge)
  |                                                       |
  +--- Agent-16 (Mobile SDK)                              |
  |                                                       |
  +--- Agent-18 (SentinelScan)                            |
  +--- Agent-19 (A2A Protocol)                            |
  +--- Agent-20 (MCP Security)                            |
  +--- Agent-23 (Multi-Tenant)                            |
  |                                                       |
  +--- Agent-17 (Deployment) <-- ALL AGENTS --------------+
              |
    +---------+---------+
    |                   |
  Agent-24            Agent-25
  (Agent Mesh)        (Edge Runtime)
    |                   |
    +---------+---------+
              |
       Master Validator

Maintained by the Archon Orchestrator. Updated as phases complete.