Agent Fabric
Orchestrate intelligent,
multi-agent systems
A unified control plane where AI agents are registered, discovered, brokered, governed, and observed — the central nervous system for your enterprise AI ecosystem.
Core Modules
Six modules. One agent control plane.
Agent Fabric ships as a single console — every module below is a working surface in the platform, not a roadmap item.
Agent Registry
The system of record for every agent in your organization. Register, version, and govern the lifecycle of internal workflow agents and external A2A agents alike.
Agent Broker
Task orchestration and execution. Submit a goal and the broker routes it to the best-suited agents, streams progress live, and keeps the full message and artifact trail.
Discovery
Scanners continuously sweep your estate and public registries for agents, then hand them to a review queue so nothing enters production unapproved.
Governance
Policies, audit trails, and an automated testing center. Enforce authentication, data governance, cost control, and compliance at every enforcement point.
Visualizer
Distributed tracing, network topology, artifact lineage, and cost analytics — real-time visibility into how your agents actually behave in production.
Protocols
First-class A2A and MCP tooling — a live playground, a tool explorer, workflow-to-agent sync, and a published master AgentCard for external discovery.
A2A Protocol
A new era of
agent interoperability.
FlowGenX AI implements the A2A (Agent-to-Agent) protocol — an open standard for agent interoperability. Agents discover each other via AgentCards, exchange tasks and artifacts, and collaborate across vendors and frameworks.
- AgentCards for capability discovery — agents advertise skills, input/output modes, and auth schemes
- Full task lifecycle — pending, working, input_required, completed, failed, canceled
- Target a specific agent or leave it blank and let the broker auto-route by capability
- Multi-modal parts — text, structured data, files, and images in a single message
- Streaming and push notifications over HTTP, SSE, JSON-RPC 2.0, and WebSocket
- Complements MCP — A2A for agent-to-agent, MCP for tools, resources, and prompts
Client and Remote agents exchange tasks and artifacts via the A2A protocol
Agent Onboarding
A guided path from endpoint to governed agent
A six-step wizard captures everything the fabric needs to route, secure, and audit an agent — or import an existing AgentCard from a URL, pasted JSON, or an uploaded file and skip straight to review.
Modalities
Execution Modes
Protocols
Authentication
Policy Tags
Cost Tier
Discovery
Find the agents
you didn't know you had.
Agent sprawl is real. Configure scanners against your cloud accounts, internal networks, and public registries — they run on an interval you set and drop everything they find into a review queue.
- Four scanner types — A2A, AWS Bedrock, Vertex AI, and public registries
- Per-scanner seed URLs, scan intervals, and enable/disable toggles
- Trigger an on-demand scan and watch the results land live
- Nothing goes live unreviewed — approve or deprecate each discovery
A2A Protocol
Crawls seed URLs for /.well-known AgentCards across your estate.
AWS Bedrock
Enumerates Bedrock agents and aliases in your AWS accounts.
Vertex AI
Discovers reasoning engines and agents in Google Cloud projects.
Public A2A Registry
Pulls published agents from public A2A registries.
Broker Routing
Single, sequential, parallel, or fallback
Submit a goal with constraints, success criteria, priority, and permitted agent scopes. The broker records its routing decision — which agents it chose, why, and which policies it evaluated — on every single task.
Single
The broker picks the one best-matching agent for the goal and hands the task straight to it.
Sequential
Agents execute in a defined chain. Each agent receives the artifacts of the previous one — perfect for staged pipelines.
Parallel
Multiple agents run simultaneously on different sub-tasks. The broker merges their artifacts when all complete.
Fallback
A prioritised chain of candidates. If the primary agent errors or times out, the broker retries down the list automatically.
Protocol Tooling
Test agent reasoning interactively
Analysing 48 open opportunities…
7 deals with no activity for 14+ days
ID: task_8f2c41ae
Goal: Flag at-risk deals in the Q3 pipeline
Summarise support backlog
Draft renewal outreach
Sync Netsuite invoices
A2A Playground
Send real A2A messages to any approved agent, or let the broker auto-route. Attach JSON context and constraints, watch task status live, and cancel mid-flight.
MCP Tools Explorer
Browse every MCP tool, resource, and prompt available to your agents. Inspect input schemas and execute a tool with test arguments right from the console.
Workflow Sync
Publish any FlowGenX workflow as an A2A agent. Define input/output schemas in a JSON editor and the fabric generates and registers the AgentCard.
AgentCard Discovery
Your tenant publishes a master AgentCard aggregating every approved agent's skills — copyable, exportable, and consumable by any external A2A client.
Governance
Policy at every enforcement point
Policies & Enforcement
Build rule-based policies with conditions, priorities, and actions. Every task the broker routes is evaluated against them before an agent ever sees it.
Policy Types
- Enforcement points — ingress, egress, inline, or all
- Actions — allow, deny, warn, redact, or audit
- Scope policies to specific agents or task types
- Versioned policies with violation severity tracking
Agent Testing Center
Generate synthetic test suites against an agent's declared capabilities, run them, and score the results — before that agent touches production traffic.
Generated Case Types
- Accuracy, safety, and completeness scored per test case
- Automatic PII-leak detection across every response
- Hallucination and off-topic flags with evaluation notes
- Pass/fail summary with latency percentiles per run
Audit Logs
Every action across the fabric is recorded and filterable.
Event Source
Registry · Broker · Gateway · Governance · Visualizer
Actor Type
User · Agent · System · Service
Resource
Agent · Task · Policy · Artifact · Message
Result
Success · Failure · Partial
Visualizer
Know exactly what your agents did, and what it cost
Distributed Traces
Span-level traces across every service a task touched, with parent/child trees, span events, and error counts.
Agent Network Topology
A live graph of agent-to-agent communication — call counts, average and max latency, and per-edge error rates.
Artifact Lineage
Trace any artifact back through the agents that produced, transformed, and consumed it.
Cost Analytics
Spend broken down by agent, tenant, and day — with daily averages, monthly estimates, and trend direction.
Performance Alerts
Threshold alerts on latency, error rate, cost, and availability, with acknowledgement and resolution tracking.
Debug Replay
Replay a task's full event timeline event-by-event, with play, pause, and variable speed, to find exactly where it went wrong.
Capabilities
Everything you need to run production agents
Agent Registry
Versioned AgentCards with draft, approved, deprecated, and suspended lifecycle states.
Capability Routing
Route by declared capability, preferred agents, exclusions, and permitted scopes.
Human-in-the-Loop
Tasks can pause for text, choice, confirm, or file input over a live WebSocket.
A2A + MCP
A2A for agent-to-agent collaboration, MCP for tools, resources, and prompts.
Endpoint Auth
API key, bearer token, mutual TLS, and OAuth 2.0 per registered endpoint.
Model Agnostic
Any provider or framework — agents interoperate through open protocols, not SDKs.
Health Monitoring
Per-endpoint health checks with latency, error detail, and needs-action flags.
JSON-First Editing
Inspect, edit, and export any AgentCard as raw JSON in an in-console editor.
Multi-Tenant
Global, org, and tenant scoping with per-environment isolation and data residency.
Orchestrate intelligent,
multi-agent systems at scale.
See how FlowGenX AI agents are registered, routed, governed, and traced — live demo tailored to your use case.