Roadmap¶

Pure orchestration library + Optional framework components

Current Status (v1.x)¶

Core orchestration is stable with 800+ tests. Key capabilities: - Task orchestration with dependency trees and priority execution - 12+ executors (REST, WebSocket, gRPC, SSH, Docker, CrewAI, LiteLLM, MCP) - Multi-protocol API (A2A, MCP, JSON-RPC) - Built-in scheduler (internal polling + external gateway) - CLI and ConfigManager - DuckDB + PostgreSQL storage

Near-term¶

Distributed Core¶

Enable multi-node deployments with centralized coordination. See design doc for details.

• Node registry with health checks
• Task leasing with automatic expiry
• Placement constraints (executor type, resources)
• PostgreSQL-based coordination

Protocol Abstraction¶

Unified interface for protocol adapters. Phase 1 complete:

• Capabilities registry (api/capabilities.py) as single source of truth for all 15 operations
• A2A adapter simplified to agent-level actions only (execute, generate, cancel)
• MCP adapter auto-generates 15 tools from the registry
• Native API (POST /tasks) as primary programmatic interface
• Method discovery endpoint (GET /tasks/methods)

Remaining:

class ProtocolAdapter(Protocol):
    async def handle_execute_request(self, request: dict) -> dict: ...
    async def handle_status_request(self, request: dict) -> dict: ...

GraphQL Adapter¶

Optional strawberry-graphql based adapter for querying task trees.

Mid-term¶

Future¶

Not Planned¶

These are application-level concerns, not orchestration:

• User Management / Auth / RBAC
• Multi-Tenancy
• Audit Logging (use observability hooks)
• Secret Management (use Vault, AWS Secrets Manager)
• Dashboard UI (separate project: apflow-webapp)

Completed¶