[DOCS] Generate the ADK provider tables from the schema (task generate)

[inference-gateway-maintainer]

Context

Follow-up to #345 (which added NVIDIA to rust-adk.md by hand) and the generator work in #339.

Onboarding a provider today is a single task generate for the gateway pages (supported-providers.md, configuration.md), but the two ADK pages carry a hand-maintained "Switching providers and models" table that the generator never touches:

• rust-adk.md:213
• typescript-adk.md:1092

Because they are hand-authored, they drift:

• typescript-adk.md predates NVIDIA, MiniMax, and Moonshot - none appear in its table.
• The NVIDIA row added by hand in docs(rust-adk): add NVIDIA provider to Rust ADK docs #345 renders Nvidia, while the generated gateway docs render NVIDIA (from the displayName override). Same provider, two spellings.

Goal: onboarding a provider should be schema + scripts/provider-overrides.json + task generate, with zero hand-editing of the ADK tables.

The table to generate

Both ADK pages share the same 4-column shape:

Column	Source
Provider	displayName override (this also fixes the Nvidia -> NVIDIA casing divergence)
A2A_AGENT_CLIENT_PROVIDER	provider id from the schema Provider enum
Example A2A_AGENT_CLIENT_MODEL	display-only, NOT modeled in the schema -> new field in provider-overrides.json
API key env var	{ENVUPPER}_API_KEY, with the auth_type: none (Ollama) row rendered as its "none (...)" note

Tasks

• Add a display-only field (e.g. adkExampleModel) per provider to scripts/provider-overrides.json. Backfill from the current tables (gpt-5-mini, claude-opus-4-8, nvidia/meta/llama-3.1-8b-instruct, ...). Providers absent from today's ADK tables (ollama_cloud, minimax, moonshot) need an example chosen.
• Add a renderAdkProviderTable(providers) renderer in scripts/generate-provider-docs.mjs, following the padded-markdown-table pattern already used by renderProvidersTable.
• Add <!-- GENERATED:adk-provider-table START/END (do not edit - run: task generate) --> markers around the table in both rust-adk.md and typescript-adk.md, and wire two new updateFile(...) calls in main().
• Handle the auth_type: none (local Ollama) row: its API-key cell is none (set A2A_AGENT_CLIENT_BASE_URL to your Ollama), not a *_API_KEY. Decide whether to derive that from the auth type or carry it as an override field (e.g. adkKeyNote).
• Extend the fixture test (scripts/generate-provider-docs.test.mjs) with an adk-provider-table case asserting a byte-for-byte match against both committed files (mirrors the existing five region tests).
• Update the GENERATED_FILES var in Taskfile.yml (line 6) to add rust-adk.md typescript-adk.md, so both the prettier --write step and the generate:check drift guard cover the new files.
• Update CLAUDE.md: the generator now manages four files / seven regions (add adk-provider-table in each ADK page), not "two files / five regions".

Design decisions to settle

1. Row order. The canonical order in provider-overrides.json (openai, deepseek, anthropic, cohere, groq, cloudflare, ollama, ollama_cloud, google, mistral, minimax, moonshot, nvidia) differs from the current curated ADK order (OpenAI, Anthropic, Groq, DeepSeek, ...). Generating adopts the canonical order - a visible but harmless reordering. Recommend accepting it for consistency with the other generated regions.
2. Display name for Ollama. The tables say "Local Ollama"; the displayName override is "Ollama". Either accept "Ollama" or add an ADK-specific label override.
3. New providers now appear in both tables. ollama_cloud, minimax, moonshot (and nvidia in the TS page) will render for the first time. That is the intended anti-drift outcome, but it is where the TS table visibly grows.
4. Per-provider prose stays hand-authored. The NVIDIA paragraph under the rust-adk.md table is editorial, not tabular - leave it (and any future per-provider note) outside the markers. Only the table is generated.

Acceptance criteria

• task generate rewrites both ADK tables; on a clean tree task generate:check passes (no drift).
• bun test covers the new region for both ADK files.
• Adding a hypothetical provider to the schema + overrides, then running task generate, updates supported-providers.md, configuration.md, rust-adk.md, and typescript-adk.md in one shot - no hand edits.

Out of scope

• The illustrative (e.g. openai, nvidia, ollama, groq) hint in the env-var reference table (rust-adk.md:971) - it is a non-exhaustive example, not a full enumeration, so it does not need generating.

[edenreich]

@claude Can you work on this?

[inference-gateway-maintainer]

Claude finished @edenreich's task in 13m 12s —— View job • docs/issue-346-20260706-0856 • [Create PR ➔](https://github.com/inference-gateway/docs/compare/main...docs/issue-346-20260706-0856?quick_pull=1&title=Issue%20%23346%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%23346%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

Done - ADK provider tables are now schema-generated

Opened PR #347. The "Switching providers and models" table is now a generated region in both ADK pages, so onboarding a provider is schema + scripts/provider-overrides.json + task generate with no hand-editing.

Completed

• Add issue to Roadmap board (In progress -> QA)
• Add adkExampleModel + adkKeyNote per provider to scripts/provider-overrides.json
• Add renderAdkProviderTable(providers) + validation + two updateFile calls in scripts/generate-provider-docs.mjs
• Wrap the typescript-adk.md table in GENERATED:adk-provider-table markers
• Add a generated "Switching providers and models" table + markers to rust-adk.md
• Extend scripts/generate-provider-docs.test.mjs with an adk-provider-table case for both files
• Update GENERATED_FILES in Taskfile.yml
• Update CLAUDE.md (four files / seven regions) - also AGENTS.md + README.md for accuracy
• Verify: bun test, lint:md, prettier --check, build, and generator idempotency

Design decisions settled (per the issue)

1. Row order: canonical override order adopted (harmless reorder of the TS table).
2. Ollama display name: renders Ollama from the shared displayName override (was Local Ollama) - no ADK-specific label added.
3. New rows now appear: Ollama Cloud, MiniMax, Moonshot (+ NVIDIA in the TS page). NVIDIA now renders NVIDIA in both places, fixing the Nvidia/NVIDIA split.
4. auth_type: none: carried as an adkKeyNote override on ollama (validated as required for keyless providers), rendering none (set A2A_AGENT_CLIENT_BASE_URL to your Ollama).
5. Prose stays hand-authored: only the table sits inside the markers.

Heads-up on repo state

The current rust-adk.md had no provider table - it was dropped in the 0.8.0 docs refresh, so the issue's rust-adk.md:213 reference and its editorial NVIDIA paragraph were stale. I re-added the section as a generated region (with a short Rust-flavored intro/closing) to meet the acceptance criteria that task generate update all four files. Worth a glance in review to confirm the placement (under Quick start, after the AI-powered server example) reads well.

Verification

• bun test -> 8 pass (6 existing + 2 new ADK cases)
• bun run lint:md -> clean
• bunx prettier --check . -> clean (generator output is prettier-stable)
• bun run build -> VitePress build succeeds
• Generator idempotent; supported-providers.md / configuration.md unchanged by the override additions
• branch docs/issue-346-20260706-0856 - PR #347

[inference-gateway-releaser]

🎉 This issue has been resolved in version 0.9.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀