← All Products
⚙️ WORKFLOW AUTOMATION STACK

Workflow Automation
By AI

Two layers of automation working together — a protocol-native workflow engine for durable multi-agent orchestration, and a task automation layer that lets you define, schedule, and monitor AI routines in plain language.

Section 1
🔧 TradeVoice Workflow Engine (TWE)

Workflow Engine

TWE is the orchestration layer that sits above the MCP + A2A protocol stack. It provides durable execution, multi-tenant isolation, and cost-aware routing — natively speaking both protocols without adapter layers.

TWE ENGINE A2A v1.0 MCP 2026-07 TRANSPORT STORAGE Durable Execution Cost Routing Tenant Isolation YAML DSL Observability Workflow Orchestration · Retry/Circuit Breaker · Model Router Agent-to-Agent Coordination Agent Cards · Task Lifecycle · Peer Delegation · Push Notifications Agent-to-Tool Communication Stateless Tool Discovery · Invocation · Resources · Prompts HTTP / SSE WebSocket gRPC Streamable Redis (state) CockroachDB (durable) MCP Servers (tools) A2A Agents (experts)
Workflow Execution Flow
🗓️ Schedule / Trigger 📋 YAML DSL Parsed ⚡ Step Executor (parallel) 🔌 MCP Tools 2026-07-28 🤝 A2A Agents v1.0 · JSON-RPC 2.0 🔍 Condition + Cost Check 📋 Summary Delivered
5 Design Pillars
Pillar 1
⚡ Protocol-Native
MCP for tools, A2A for agents — core primitives, not adapters
Pillar 2
🔀 Process-Aware
YAML workflows with steps, conditions, parallel branches, approval gates
Pillar 3
🧩 Domain-Composable
DomainBundles package MCP servers + A2A agents + workflow templates
Pillar 4
🔒 Multi-Tenant
Per-user credential isolation, state namespacing, execution sandboxing
Pillar 5
💰 Cost-Aware
Per-step model routing, token budgets, cost tracking, circuit breakers
Runtime Components
WorkflowEngine StepExecutor YAML DSL Parser ModelRouter CostOptimizer ToolRegistry AgentRegistry TenantContext CredentialStore ObservabilityBus CircuitBreaker RetryPolicy Redis (state) CockroachDB (durable)
Example: Stock Analysis Workflow (YAML DSL)
name: stock_analysis
version: "1.0"
schedule: "0 9 * * 1-5"  # weekdays at 09:00

steps:
  - id: fetch_signals
    type: tool
    tool: ta_trend_signals     # via MCP
    params: {symbol: "{{ input.symbol }}"}

  - id: risk_check
    type: agent
    agent: risk_analyst        # via A2A
    depends_on: [fetch_signals]
    parallel: true

  - id: equity_analysis
    type: agent
    agent: equity_analyst      # via A2A
    depends_on: [fetch_signals]
    parallel: true

  - id: summarise
    type: llm
    depends_on: [risk_check, equity_analysis]
    model_budget: low           # cost-aware routing
Section 2
⚙️ TASK AUTOMATION

Task Automation

Define your routines in plain language. The automation layer breaks them into steps, assigns AI agents, executes on schedule, monitors conditions, and delivers a structured summary — no boilerplate, no glue code.

HOW A ROUTINE RUNS
📝 Define Routine
→
🤖 AI Planner breaks into steps
→
⚡ Agents execute in parallel
→
✅ Condition checks
→
📋 Summary delivered
🗓️
Schedule & Triggers

Cron schedules, market-event triggers, and condition-based firing — routines run exactly when you need them. Pre-market, post-market, or custom intervals.

🤖
AI Step Planner

Describe a goal in plain language. The planner decomposes it into a sequence of agent tasks automatically — no workflow diagram required.

🔀
Parallel Execution

Independent steps fan out and run simultaneously across agents. Dependent steps wait for upstream results. Finish faster, not slower.

🔍
Condition Monitor

Set alert conditions — price levels, indicator thresholds, news keywords — that trigger downstream actions automatically without manual polling.

📋
Result Summaries

Every run produces a structured memo — what happened, what signals fired, what needs attention. Stored and searchable across runs.

🔌
Pluggable Agents

Connect any A2A domain expert, MCP tool, or custom agent as a step in your routine. Framework-agnostic — LangGraph, CrewAI, or plain Python all work.