LF AI & Data Sponsor Map¶
Evidence date: 2026-08-01. State: research only; no organization below has been contacted, agreed to sponsor Fgentic, or endorsed it. Issue #1626 owns staff confirmation and later outreach.
Resolve the published-policy conflict first¶
The official sources do not currently describe the same Sandbox gate:
- The pinned Project Lifecycle Document requires a Sandbox applicant to have an existing LF AI & Data member sponsor, or a new organization that joins and sponsors the application.
- The live project-services page instead says a Sandbox project must be contributed by a member and places the explicit sponsor requirement under Incubation.
- The live membership page describes Associate membership as complimentary for qualifying government bodies, nonprofits, and universities. The applicant must be a qualified representative, and the Governing Board must approve the application. The charter calls Premier, General, and Associate organizations members, but neither source says whether an Associate may sponsor or contribute a Sandbox project.
This map is therefore a targeting aid, not proof that any row can satisfy the gate. Before outreach, #1626 must obtain written LF AI & Data staff answers to three questions:
- Which published member or sponsor rule governs a new Sandbox application today?
- May an existing Associate member satisfy that rule?
- May a newly approved Associate satisfy it, and what timing or Governing Board conditions apply?
Do not open sponsor conversations until the maintainer's sequencing gates are complete: the v1.0.0 tag (#471), named external evaluators (#1671), and the launch announcement (#69).
Do not open the proposal tracked by #1627 without a confirmed valid route. The lifecycle imposes a six-month wait after a rejected stage-transition request.
Existing-member candidates¶
The live LF AI & Data member directory contained 4 Premier, 17 General, and 45 Associate members on the evidence date. Ranking favors verified A2A, MCP, Kubernetes, agent-governance, foundation-process, and sovereignty adjacency; it does not infer willingness or endorsement. Recheck both membership and technical evidence on the day outreach begins.
| Rank | Current member and tier | Verified adjacency | Why it is in this segment |
|---|---|---|---|
| 1 | IBM — Premier | IBM documents A2A interoperability and its BeeAI adapters; IBM's ACP work joined A2A. | Direct protocol overlap with Fgentic's official-SDK Matrix-to-A2A bridge and cross-organization delegation boundary. |
| 2 | Okahu — General | Okahu has already contributed Monocle as an LF AI & Data Sandbox project and publishes production AI-agent observability and evaluation. | The closest verified foundation-process experience plus direct fit with Fgentic's delegation, evaluation, token, and trace evidence. |
| 3 | Nokia — General | Nokia and Telefónica report testing A2A and MCP for agent-to-network interoperability. | Strong European, open-protocol, multi-vendor, and cross-domain trust adjacency. |
| 4 | NVIDIA — Premier | NeMo Agent Toolkit publishes A2A client/server support built on the official SDK alongside MCP. | Direct A2A plus MCP overlap and a Kubernetes-oriented agent-runtime ecosystem. |
| 5 | OpenUK — Associate | OpenUK's 2026 AI Openness report covers agentic infrastructure, MCP, open standards, and sovereignty. | Strong mission and regional fit, conditional on written confirmation that an Associate may satisfy the gate. |
| 6 | Banque de France — Associate | The bank describes its secure, sovereign AI program and reversibility requirements. | Existing Associate with unusually close public-sector sovereignty, control, and regulated-operation priorities; no A2A claim is implied. |
| 7 | AMD — General | AMD publishes an A2A multi-agent tutorial using locally served models. | Direct A2A and self-hosted-inference evidence, but primarily an infrastructure rather than collaboration fit. |
| 8 | Union.ai — General | Union documents running models and agents as pods on infrastructure the adopter owns. | Kubernetes, self-hosting, workload isolation, and data-residency adjacency. |
| 9 | Cloudera — General | Cloudera's Agent Studio publishes multi-agent, tool, sandbox, and data-sovereignty boundaries, with MCP support in its release documentation. | Enterprise agent operations and governed local-data concerns overlap, although its product surface is broader than Fgentic's bridge. |
| 10 | Databricks — General | Databricks documents multi-agent applications and governed MCP services. | Multi-agent and governed-tool adjacency; sovereignty alignment must be tested rather than assumed. |
| 11 | ZTE — Premier | ZTE publishes an agent-network architecture with A2A and MCP compatibility. | Agent communication, identity, and gateway adjacency; interoperability remains a candidate discussion, not established Fgentic compatibility. |
| 12 | Huawei — Premier | Huawei announced an open-source A2A-T SDK, registry, and orchestration center. | Agent-communication and registry adjacency. A2A-T is a telecom protocol and must not be presented as LF A2A compatibility. |
The A2A, kagent, and agentgateway communities are useful relationship paths, but project adjacency is not LF AI & Data membership. On the evidence date, the current member directory does not list most organizations named in those adjacent communities, including Solo.io and Red Hat. Never use a cached member announcement or shared Linux Foundation participation as proof of current LF AI & Data membership.
Potential new-Associate track¶
These organizations match a published legal category and have a concrete Matrix, agent-infrastructure, or sovereignty connection. That makes them research candidates for the complimentary Associate route, not pre-approved members or sponsors. LF AI & Data staff and each organization's qualified representative must confirm eligibility and interest.
| Rank | Candidate | Published basis and adjacency | Unresolved gate |
|---|---|---|---|
| 1 | DINUM | French government body; it operates the Matrix-based Tchap service and the sovereign Albert API. | Staff must confirm that the relevant government entity may join and later satisfy the Sandbox rule. |
| 2 | Eclipse Foundation | International nonprofit; Eclipse LMOS is an open, Kubernetes-based multi-agent platform. | Separate-foundation custody and sponsorship authority require explicit agreement. |
| 3 | Matrix.org Foundation C.I.C. | The nonprofit guardian of the Matrix standard is the closest protocol steward to Fgentic's adopter surface. | No AI-infrastructure interest or LF AI & Data intent is inferred. |
| 4 | TU Dresden | University operating an institution-wide Matrix chat service. | Agent-infrastructure interest and authorized representation remain unverified. |
| 5 | ZenDiS | Public-sector sovereignty organization behind the Kubernetes-based openDesk distribution with Element and Synapse. | Its GmbH legal form is not clearly within the published Associate categories; ask staff before treating it as eligible. |
Outreach playbook for #1626¶
- Wait for #471, #1671, and #69; do not spend the first-impression opportunity on a pre-release project without external evaluation evidence.
- Recheck the LF AI & Data lifecycle, services page, membership rules, charter, and member directory on the same day.
- Obtain the three written staff clarifications above before asking any candidate to sponsor or join.
- Select at least eight candidates from the now-valid route. Start with the highest-ranked existing members; use potential Associates only if staff confirms that route.
- Lead with the public demo recording and trust evidence, then the narrow product claim: the Matrix-to-A2A bridge and standalone chart. Do not pitch the reference distribution as a monolithic AI platform.
- Append the date, organization, route, evidence sent, response, and decline reason to the outreach ledger below, then link the documentation update from #1626. A non-response is not a decline until the outreach owner records its follow-up boundary; issue comments coordinate the work but are never the sole record.
Outreach ledger¶
No outreach has occurred as of the evidence date. When #1626 starts, append one row per candidate and update that row as the conversation progresses; never remove a decline or replace a non-response with an inferred outcome.
| Organization | Route | First contact (UTC) | Evidence sent | Response and outcome | Last checked (UTC) |
|---|---|---|---|---|---|
No-sponsor fallback¶
The one-quarter clock is three calendar months. It starts on the first recorded candidate outreach after the three sequencing gates and staff clarification; the outreach owner records that start date and deadline on #1626. It does not start on this document's evidence date.
If no eligible sponsor or contributing member is secured within that quarter:
- Record every outcome in the outreach ledger and stop the round; do not convert silence into an endorsement or repeatedly solicit the same organizations.
- Keep Fgentic independent and continue normal Apache-2.0 releases, public governance, security evidence, adopter support, and protocol work. Foundation acceptance is not required to ship the bridge.
- Keep the CNCF 2028 second-credential dossier dated and accurate, but do not submit it or imply a dual-hosting path.
- Accumulate post-v1 evidence: a released
v1.0.0, named external demo runs, a real adopter or pilot, and external human contributions. - Revalidate LF AI & Data policy and the member roster, then open at most one new outreach round when the evidence has materially changed. If the policy or custody fit no longer serves the project, remain independent rather than forcing a foundation outcome.
This fallback is a success condition for the research track: it protects release and adoption work from an external gate while preserving a better-evidenced future application.