WorkingAgents + xpander.ai: The Brain Gets a Body, The Body Gets a Brain

By James Aspinwall, co-written by Alfred Pennyworth (my trusted AI) — March 7, 2026, 18:30

The Core Insight

WorkingAgents has the intelligence — 50+ purpose-built MCP tools, permission-aware orchestration, CRM, task management, and human communication channels. What it doesn’t have is a framework-agnostic deployment runtime, a visual agent builder for non-developers, or a library of 2,000+ pre-built connectors.

xpander.ai has all of that. They’re the runtime and control plane for building, deploying, and shipping AI agents across any framework, any cloud, any model. What they need is the kind of rich, domain-specific business tools that turn generic agents into useful ones.

xpander.ai’s tagline captures it perfectly: “Your AI Agent Has a Brain. xpander.ai Gives It a Body.” WorkingAgents provides the business intelligence those bodies need to actually do work.

What xpander.ai Brings

xpander.ai is a full-stack agent platform founded by Ran Sheinberg and David Twizer in 2024, headquartered in Tel Aviv. With $3M in funding from Emerge Ventures, Samsung NEXT Ventures, and SeedIL Ventures, a lean team that hit $1.1M revenue in 2025, and an open-source runtime under Apache 2.0, they’ve built the infrastructure layer that sits between the LLM brain and the production environment.

AgentOS — The Runtime

AgentOS is xpander’s production runtime for AI agents. It handles everything below the application layer:

AgentOS deploys in three modes:

  1. Managed serverless — xpander’s cloud, auto-scaling, no infrastructure to manage
  2. Embedded containers — your container, your dependencies, full runtime control, GPU support
  3. Self-hosted Kubernetes — your VPC, your cluster, complete data isolation

The Workbench — No-Code Agent Builder

The Workbench is a visual control plane where operations teams and domain experts build agents without writing code:

This is the layer WorkingAgents doesn’t have — a visual builder for non-developers.

Connector Hub — 2,000+ MCP-Ready Tools

The Connector Hub is a repository of over 2,000 pre-built connectors for enterprise systems:

Each connector runs as an MCP server, making them discoverable and invocable through the standard protocol.

Agent Graph System (AGS) — Guided Orchestration

xpander’s Agent Graph System is a directed graph that controls which tools an agent can access at each step. Instead of presenting all available tools to the LLM and hoping it picks correctly, AGS restricts the option set to only tools that align with the current task context.

This significantly reduces out-of-sequence function calls — one of the most common failure modes in multi-tool agents. The LLM doesn’t need to micromanage each sub-tool; AGS delegates work through controlled sequences.

MCP-Native Architecture

xpander.ai is deeply invested in MCP:

Framework Agnostic

One unified API works across five major agent frameworks:

And supports all major models: Claude, GPT-4, Gemini, Llama, DeepSeek.

Pricing

What WorkingAgents Brings

WorkingAgents (“The Orchestrator”) is an Elixir OTP platform with production-grade business tools:

Where the Synergy Lives

1. WorkingAgents as an MCP Server in xpander’s Connector Hub

This is the most direct integration. WorkingAgents is already an MCP server exposing 50+ tools. xpander’s Connector Hub is a library of MCP servers. WorkingAgents becomes a connector in the Hub.

Any agent built on xpander — regardless of framework (LangChain, OpenAI, Agno, Google ADK, AWS Strands) — could immediately access:

Today, xpander’s Connector Hub has Microsoft 365, Salesforce, Jira, Slack. Adding WorkingAgents’ business operations tools creates a new category — not a SaaS integration, but a full business intelligence layer accessible to any agent on any framework.

2. Agent Graph System + WorkingAgents Tool Chains

WorkingAgents has 50+ tools, but agents don’t always pick the right sequence. xpander’s Agent Graph System (AGS) solves exactly this problem.

Consider a “process new lead” workflow using WorkingAgents’ CRM tools:

Without AGS — the LLM sees all 50+ tools and must figure out the sequence on its own. It might call nis_create_contact before checking if the contact already exists via nis_search, or skip nis_log_interaction entirely.

With AGS — the graph defines the sequence:

Step 1: nis_search (check if contact exists)
  → Found: nis_get_contact → Step 3
  → Not found: Step 2

Step 2: nis_create_contact (create new record)

Step 3: nis_log_interaction (record the interaction)

Step 4: task_create (create follow-up task)

Step 5: task_link (link task to contact)

At each step, AGS only exposes the relevant tools. The LLM can’t accidentally jump ahead or skip steps. This transforms WorkingAgents’ tool library from “50 tools the LLM must navigate” to “guided workflows the LLM follows reliably.”

For consulting clients, this is the difference between “the agent sometimes creates contacts without logging the interaction” and “the agent follows the defined process every time.”

3. Visual Agent Builder for WorkingAgents’ Tools

WorkingAgents is powerful but developer-centric. Building a new agent workflow requires Elixir/IEx knowledge, understanding of MCP tool names and parameters, and familiarity with the permission system.

xpander’s Workbench changes this entirely:

This dramatically expands WorkingAgents’ addressable market. Today, only developers can build on the platform. With xpander’s Workbench, domain experts can.

4. Framework-Agnostic Deployment

WorkingAgents’ MCP server is framework-agnostic by protocol, but in practice it’s an Elixir application that runs on a specific server. xpander adds deployment flexibility:

WorkingAgents’ business logic stays in Elixir/OTP. xpander handles the last-mile framework and cloud adaptation. The client picks their preferred stack; the tools work everywhere.

5. Connector Hub Generation for Custom Client APIs

xpander’s Connector Hub can generate MCP servers from any OpenAPI spec. For WorkingAgents’ consulting engagements, this is a force multiplier:

This eliminates the custom integration work that slows consulting engagements. Instead of writing bespoke code for each client’s systems, generate connectors and wire them into the agent graph.

6. Multi-Agent Orchestration at Scale

xpander’s Chat interface autodiscovers deployed agents and can schedule tasks across them. Combined with WorkingAgents’ A2A protocol, this creates a multi-layer orchestration architecture:

A user asks via xpander Chat: “What’s the status of the Johnson deal?” The Chat discovers a sales agent, which calls WorkingAgents’ nis_get_company, nis_list_contacts (filtered by company), and task_query (tasks linked to that company). The response aggregates CRM data, open tasks, and last interaction — all orchestrated through the multi-agent mesh.

7. Self-Hosted Deployment for Consulting Clients

WorkingAgents runs on Elixir/OTP, which some enterprise clients may find unfamiliar. xpander’s self-hosted Kubernetes deployment provides a familiar operational model:

The consulting pitch becomes: “You deploy on your own Kubernetes cluster, using standard DevOps practices. The AI agent runtime is managed by xpander. The business tools are powered by WorkingAgents. Everything stays in your infrastructure.”

The Gap Analysis

WorkingAgents Gap xpander.ai Solution
No visual/no-code agent builder Workbench — non-developers build agents visually
No guided tool sequence control Agent Graph System — directed graphs prevent out-of-sequence calls
No framework-agnostic deployment One API for Agno, LangChain, OpenAI, Google ADK, AWS Strands
No serverless/container deployment options Managed serverless, embedded containers, self-hosted K8s
No connector generation from API specs Connector Hub generates MCP servers from OpenAPI docs
No CI/CD for agent versions AgentOS with versioning, testing, and rollback
Limited enterprise deployment infrastructure Helm Charts, Docker, SSO, SLA, managed onboarding
xpander.ai Gap WorkingAgents Solution
Need rich business-domain tools 50+ MCP tools for CRM, tasks, content, scheduling, monitoring
Need relationship intelligence layer NIS with contacts, companies, pipeline, follow-ups, activity logging
Need human communication channels WhatsApp bridge, real-time WebSocket chat
Need permission-aware tool execution Capability-based access control on every tool call
Need agent-to-agent protocol beyond MCP Google A2A protocol for cross-platform skill discovery
Need fault-tolerant runtime for business logic Elixir OTP supervision trees with hot code reloading
Need real business outcome data Pipeline tracking, task completion, contact follow-up metrics

Partnership Model

Phase 1: Connector Integration (Week 1-2)

Publish WorkingAgents as a connector in xpander’s Connector Hub:

Phase 2: Joint Reference Architectures (Weeks 3-6)

Build and document reference architectures for consulting:

Phase 3: Consulting Delivery Model (Months 2-4)

Position the joint stack for consulting engagements:

Revenue Model

Why xpander Over Other Agent Platforms

xpander occupies a unique position relative to the other platforms in this partnership series:

Platform What It Provides Relationship to WorkingAgents
Arize AI Observability — traces what happened Monitoring layer
Deepchecks Evaluation — scores whether it was good Quality layer
Distributional Analytics — discovers behavioral patterns Intelligence layer
Lyzr.ai Enterprise agents — domain-specific automation Vertical partner
xpander.ai Runtime + deployment — the infrastructure to run agents anywhere Infrastructure partner

xpander is the only partner in this series that provides the deployment and runtime infrastructure WorkingAgents lacks. The others add observability, evaluation, analytics, or domain depth. xpander adds the ability to run WorkingAgents’ tools on any framework, any cloud, with any model, through a visual builder that non-developers can use.

Three specific differentiators:

1. Agent Graph System. No other platform in this series offers guided tool sequence control. For WorkingAgents’ 50+ tools, AGS is the difference between “agents sometimes work” and “agents reliably follow defined processes.” This is critical for consulting clients in regulated industries.

2. Connector Hub generation. The ability to create MCP servers from any OpenAPI spec means consulting engagements can connect client-specific systems in hours, not weeks. Combined with WorkingAgents’ 50+ tools, every client deployment starts with a rich tool library and expands rapidly.

3. Framework agnosticism. WorkingAgents’ MCP server works with any MCP client, but xpander makes this practical by providing one unified API across five major frameworks. Clients don’t need to learn MCP — they use their preferred framework, and xpander handles the protocol translation.

Conclusion

The partnership opportunity with xpander.ai is infrastructure-level, not feature-level. xpander doesn’t compete with WorkingAgents’ business tools — it provides the deployment fabric that makes those tools accessible to any agent, on any framework, in any cloud, built by anyone.

For the consulting firm, xpander solves three practical problems:

  1. “Our team doesn’t know Elixir” — doesn’t matter. Build agents against WorkingAgents’ MCP tools using LangChain, OpenAI, or whatever framework the client’s team already knows. xpander’s Workbench lets non-developers build visually.

  2. “How do we ensure agents follow the process?” — Agent Graph System. Define the tool sequence as a directed graph. The agent can’t skip steps or call tools out of order.

  3. “Where does this run?” — Anywhere. xpander’s managed cloud, the client’s Kubernetes cluster, or a self-hosted Docker deployment. AgentOS handles runtime, state, secrets, and CI/CD.

WorkingAgents is the brain — the business intelligence, CRM, task management, and communication layer. xpander.ai is the body — the runtime, deployment, and infrastructure that lets that brain act across any environment. The MCP protocol connects them natively. The integration path starts with publishing WorkingAgents as a connector in xpander’s Hub — a day’s work that opens every framework and every cloud to WorkingAgents’ 50+ tools.

Sources: