[CNCF LFX Proposal] Meshery BYOM: Adapter for AI and LLMs

[hortison]

CNCF Project

Meshery

Term

2026 Term 3 (Sep-Nov)

Program Name

BYOM: Adapter for AI and LLMs

Program Description

Description

Meshery's AI Adapter track aims at natural language to infrastructure: a user describes architectural intent and Meshery renders a valid Design in Kubernetes clusters and Meshery extensions like Kanvas. Early implementations configured providers through startup environment variables such as MESHERY_AI_OPENAI_API_KEY, which does not survive contact with real deployments. That approach is per-server rather than per-user, it cannot express more than one provider, and it puts secrets somewhere Meshery already has a better home for them. Meshery has first-class Connections and Credentials with schema-driven registration, and server-side CRUD for model-provider connections has already landed. What is missing is the complete path from a user supplying their own key to a generated, validated Design on the canvas.

This project makes AI providers bring-your-own and user-owned. The mentee implements provider Connections and Credentials spanning cloud and local inference (OpenAI Codex, Anthropic Claude, GitHub Copilot, Azure OpenAI, AWS Bedrock, Vertex AI, Ollama, LocalAI), a Create New Connection experience in Meshery UI, provider health and readiness checks, and the generation path itself: prompt, to compact model and schema context, to provider call, to candidate Design, to validation, to a deployable workload in a Kubernetes cluster, and a reviewable result in Kanvas. Secrets live only in Credentials. They are never returned to clients, never written into prompt context or generated designs, and never appear in logs or events. Generation produces a candidate for human review and never deploys automatically, which keeps Meshery UI as the review surface and keeps the feature safe to enable by default.

Expected Outcomes

• Connection and Credential support for at least four provider kinds spanning hosted and local inference, registered through Meshery Models and manageable from both Meshery UI and mesheryctl.
• A Create New Connection wizard plus provider health and readiness checks surfaced in UI and CLI, with defined status, error, and event semantics and operationId correlation.
• End-to-end natural language to Design generation using user-supplied credentials, returning a reviewable candidate Design or structured validation errors rather than an automatic deployment.
• A demonstrated provider swap between a hosted frontier model and a local model with no code changes, plus secret redaction across logs, events, and API responses verified by tests.
• Documentation covering provider setup, the credential contract, data handling and privacy posture, and an AI production checklist added to the system check guide.

Technologies

Golang, Javascript, React, REST, Frontier model provider APIs / LLM provider APIs

Skills same as Technologies?

• Yes, the required skills are the same as the technologies listed above.

Required/Desirable Skills

Kubernetes

Mentors

Lee Calcote | @leecalcote | leecalcote@gmail.com | leecalcote
James Horton | @hortison | james.hortison2337@gmail.com | hortison

Upstream Issue URL

meshery/meshery#20994

Application Prerequisites

• Resume
• Cover Letter
• School Enrollment Verification
• Participation Permission from school or employer
• Coding Challenge
• Custom Prerequisite (fill in details below)

Coding Challenge URL

No response

Custom Prerequisite Name

Participation is Key

Custom Prerequisite Description

1. Join the community (https://slack.meshery.io).
2. Join the weekly newcomers meeting (https://meet.meshery.io) and learn about the project.
3. Get a feel for Meshery's capabilities in the Playground (https://play.meshery.io)
   1. Create a design.
   2. Receive a badge (from https://meshery.io/community/recognition/).
      Optionally:
   3. Add badge to your GitHub profile.
   4. Publish your design the Catalog.
4. Ascertain whether your passions and skills align.

Custom Prerequisite — File Upload

• Yes — completion of this task requires the mentee to submit a file.

---

LFX program: CNCF - Meshery: BYOM: Adapter for AI and LLMs (2026 Term 3)

[github-actions]

🤖 LFX Proposal Validation

LFX program title (59/100 chars):

CNCF - Meshery: BYOM: Adapter for AI and LLMs (2026 Term 3)

✅ All checks passed.

⚠️ 1 warning(s):

• LFX LFID/email verification is not yet automated. Please manually confirm mentor emails are registered on https://mentorship.lfx.linuxfoundation.org.

✅ Passed checks

• CNCF Project: Meshery
• Term: 2026 Term 3 (Sep-Nov)
• Program Name length: 29/70 chars available
• Program Description: 2675/3000 chars
• Technologies: Golang, Javascript, React, REST, Frontier model provider APIs / LLM provider APIs
• Skills: Kubernetes
• Mentors: 2 listed (primary: Lee Calcote)
• Upstream Issue URL: Feature: Adapter for AI and LLMs meshery/meshery#20994
• Custom Prerequisite: name 20/20 chars, description 469/500 chars
• Quota: 2/4 proposals for Meshery in 2026 Term 3 (Sep-Nov)

What happens next

1. Maintainer approval — ✅ recorded; proposer @hortison is a project maintainer.
2. Mentor confirmation — ⏳ 1 of 2 confirmed; @hortison auto-counted as proposer. Awaiting /confirm from @leecalcote.
3. CNCF approval — ⏳ a CNCF admin comments /cncf-approve to finalize.

Proposal process guide

---

Re-run by editing the issue.

[nate-double-u]

Thanks for this @hortison, a note, I don't think the LFX platform supports markdown in the custom prerequisite description field, so you may want to either strip it out to plain-text, or fold some of the pre-req description into the main program description.