Agents that understand telecom.
Trusted autonomy for production.
Autonomous Networks Level 4 is achievable. The harder problem is keeping it there — safely, trusted, and business-focused — across multi-vendor agent estates.
Agent Fabric is the runtime where AI agents from any vendor discover each other, coordinate work, follow the operator’s process, and prove what they did. It is built on A2A-T, the telecom extension to the Linux Foundation Agent-to-Agent protocol, specified in TM Forum IG1453 and contributed back as open source.
A2A-T replaces unpredictable natural-language agent chatter with structured, auditable, low-latency exchanges across three profiles: Task-T, Event-T, and Negotiation-T. Around the runtime sits an Agent Registry, a Prompt Registry, a Governance Agent, and an operator BPMN-to-Task-T translation layer that turns years of encoded institutional process into agent-executable work.
Agent Fabric contributes a Digital Twin implementation of the Service Fault Management High-Value Scenario, plus two net-new HVS — Enterprise Complaint Handling and Enterprise Service Fulfillment — back to TM Forum’s IG1339 v2.4.0.
85% of CIOs say explainability gaps have already blocked AI from production. The industry needs an open protocol, an open runtime, and a governance fabric — not another agent product.
Operators are under board pressure to prove AI ROI, with regulators requiring traceability within 12 months. A2A-T gives operators their own processes as the foundation — workflows compiled into agent tasks. Every decision produces an evidence chain.
74% of CIOs regret at least one AI vendor selection in the past 18 months. The root cause is lock-in. A2A-T, grounded in Linux Foundation open source, means any vendor’s agent registers once and interoperates across the entire fabric.
82% of CIOs say agents are built faster than IT can govern them. Consultants need a standards-backed governance blueprint deployable as an overlay on existing OSS/BSS — with no replatforming.
Agent Fabric, OpenAN, A2A-T — the three terms get used interchangeably. They aren’t. Each layer does a distinct job, and each is open.
How A2A-T, the registries, the process layer, and the governance layer come together to deliver trusted autonomy at production scale.
A2A-T runtime: Task-T, Event-T, Negotiation-T. Structured, low-latency, auditable inter-agent communication. Open sourced via Linux Foundation.
Prompt Registry with structured, versioned prompts. Every agent gets the right telecom context every time. Deterministic where it matters.
Agent Registry with Agent Cards, skills, and onboarding trust. Agents from any vendor are discoverable and coordinated without bespoke integration.
BPMN-to-Task-T translation. Operator processes — the institutional knowledge encoded over years — become agent-executable work. Governance stays with the operator.
Audit, Governance Agent, agent evaluation, and explainability are fabric-level capabilities. Not bolt-ons. Autonomy you can prove.
The TM Forum AN Level taxonomy. Agent Fabric targets the L3→L4 transition where multi-agent cross-domain coordination becomes mandatory and intent meets agents.
Human operators perform every task. No automation.
Tools assist humans. Operators decide and act.
Systems execute closed loops under defined conditions. Humans supervise.
Self-configuring under specified conditions. Predictive operation begins.
Intent-driven, agent-coordinated, cross-domain. Operators set policy. Agents execute.
Agent Fabric targetSelf-generated intent. The network reasons about its own goals.
The integrated demo shows the full fabric in motion. Operational-proof scenarios anchor outcomes against TM Forum High-Value Scenarios. Capability demos isolate each component.
Autonomous fault resolution spanning multiple coordinated agents across RAN and Transport domains. End-to-end, agent-coordinated, no human handoffs.
SLA breach triggers cross-layer resolution. Business-layer and service-layer agents collaborate to root-cause, remediate, and close with structured outcomes including SLA credit.
Internal and external agents coordinate autonomous provisioning across access and transport layers. Compressed timelines, third-party agents in the loop, no manual orchestration.
Telecom-grade agent communication
Open-source orchestration runtime
Multi-agent resolution on Digital Twin
Operator BPMN, deterministic execution
Trust scored before deployment
Agent decisions traced
Cross-domain visibility
Constraints honoured at runtime
TM Forum SID as OWL knowledge graph, queryable via SPARQL and MCP
AN use cases as knowledge graph, MCP-accessible
A2A-T extends the Linux Foundation Agent-to-Agent protocol with three telecom profiles. It carries existing intent vocabulary — TMF921, 3GPP TS 28.312 — rather than replacing it. Specifications are open. The proto is on GitHub. The realisation guide is published by TM Forum.
Phase III contributes upstream to the Linux Foundation A2A project and ships its core runtime, registries, protocol SDK, and ontologies as public repositories. Core repos go public around DTW 2026 as they graduate.
Agent Fabric is grounded in TM Forum’s published asset family for Autonomous Networks, Intent, and Agentic Trust. It does not invent in isolation.
| Asset | Role | Relationship |
|---|---|---|
| IG1453 v2.1.0 | A2A-T Specification | Authored |
| IG1453A v1.1.0 | A2A-T Realisation Guide | Authored |
| IG1339 v2.4.0 | AN HVS Map | Contributing · 1 Twin + 2 net-new HVS |
| GB1087 v2.0.0 | Agentic Interaction Security in Telecoms | Aligned to |
| IG1251 | AN Reference Architecture | Anchor |
| IG1251C | AN Agentic Architecture | Anchor |
| IG1251D | Agent Anatomy | Anchor |
| IG1251E | Multi-Agent Systems | Anchor |
| IG1358 | Intent-Based Operation User Guide | Referenced |
| IG1218F | AN Framework | Referenced |
| IG1457 | AN Journey | Referenced |
| TMF921 | Intent Management API | Carried |
| TMF641 / 621 / 642 | Service Management APIs | Carried |
| TIO | TM Forum Intent Ontology | Carried |
| 3GPP TS 28.312 | Intent NRM (Rel-19) | Carried |
| LF A2A | Base Protocol | Extends & Contributes |
Agent Fabric is a multi-stakeholder project. Champions shape direction. Vendors build the runtime. The University of Milan anchors the academic research underpinning the Trust arc.
The team’s work spans continuous trust certification, Trustworthy AI, behavioural trust evaluation, and verification of complex digital ecosystems — including LLM-based and regulated systems. Provides the academic foundation underpinning Agent Fabric’s Trust arc.
Profile at UNIMI ↗IG1453 v2.1.0 defines the A2A-T protocol. IG1453A v1.1.0 is the realisation guide. Both are published through TM Forum.
C26.0.910 is the third phase of the Agent Fabric Catalyst. Live demonstration at DTW Ignite 2026 in Copenhagen.
For collaboration enquiries, partnership opportunities, or to join the community, reach the project lead.
The A2A-T extension proposal is live on the Linux Foundation A2A project. Core repositories graduate around DTW 2026.