Open the HuggingFace + Ollama "Use this model" listing PRs (protolabs-agent is on PyPI)

[mabry1985]

Context

protolabs-agent 0.101.0 is live on PyPI now (uv tool install protolabs-agent → the protoagent command), so the two community-listing pieces from #1833 (HuggingFace local-app) and #1834 (Ollama Community Integrations) finally have a real install target and are unblocked. Both snippets were validated end-to-end against real Ollama (Ornith-1.0-9B → COMPLETED).

Deferred until after more app marketing + demo recordings. This issue is just "open the two external PRs" — public PRs to third-party repos under protoLabs' name.

⚠️ The PyPI name is protolabs-agent, not protoagent (the plain name is similarity-blocked by an unrelated proto-agent). The CLI command is still protoagent. Install line in both snippets is uv tool install protolabs-agent.

1. HuggingFace — PR to huggingface/huggingface.js

File packages/tasks/src/local-apps.ts. Add snippetProtoagent near snippetHermesAgent; insert the entry alphabetically between openclaw and pi.

const snippetProtoagent = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
	const modelId = isMlxModel(model) ? model.id : `${model.id}${getQuantTag(filepath)}`;
	const serverStep = getLocalServerStep(model, filepath);
	return [
		serverStep,
		{
			title: "Point protoAgent at the local server",
			setup: ["# Install protoAgent:", "uv tool install protolabs-agent"].join("\n"),
			content: [
				"# Point protoAgent at the local server:",
				`protoagent model use --base-url http://127.0.0.1:8080/v1 --model ${modelId}`,
			].join("\n"),
		},
		{ title: "Run protoAgent", content: "protoagent up" },
	];
};

// in LOCAL_APPS, between "openclaw" and "pi":
protoagent: {
	prettyLabel: "protoAgent",
	docsUrl: "https://github.com/protoLabsAI/protoAgent/blob/main/docs/guides/cli.md#point-at-a-local-model",
	mainTask: "text-generation",
	displayOnModelPage: isToolCallingLocalAgentModel,
	snippet: snippetProtoagent,
},

Validated details: gate is isToolCallingLocalAgentModel (a non-tool model 400s does not support tools); model use strips the :{{QUANT_TAG}} placeholder; launch verb is protoagent up (bare protoagent prints help). Review ≈ 1–2 weeks.

2. Ollama — PR to ollama/ollama

File README.md, under ### Terminal & CLI (one terse line, matching neighbors):

- [protoAgent](https://github.com/protoLabsAI/protoAgent) - LangGraph agent runtime with an operator console, A2A, and drop-in plugins; runs any Ollama model via its OpenAI endpoint

Acceptance

• huggingface.js PR opened (fork + local-apps.ts entry above)
• ollama/ollama PR opened (fork + README line above)
• (nice-to-have) "Run with protoAgent" section added to our published HF GGUF/MLX quant cards (needs tags: conversational + a GGUF whose chat_template includes tools)
• Link the PRs back on D5: register protoAgent as a HuggingFace "Use this model" local-app (ADR 0075) #1833 / D5: add protoAgent to Ollama Community Integrations (ADR 0075) #1834; close those when merged

Refs: #1833, #1834. The plumbing that unblocked this: PyPI publish via #1984 / #1988 / #1989.

[mabry1985]

Absorbing #1833 (HuggingFace local-app listing) and #1834 (Ollama Community Integrations) — this issue IS those two D5 items, now unblocked since protolabs-agent is on PyPI. Two listing PRs, one ticket.

[mabry1985]

Both PRs opened:

• huggingface.js: Add protoAgent local-app snippet huggingface/huggingface.js#2372 — verified against the CURRENT file (the issue's "insert alphabetically between openclaw and pi" anchor had drifted; the file isn't strictly alphabetical and openclaw is now the last entry, so the new protoagent entry was placed right after it instead). Added protoagent/protoagent - mlx tests mirroring the hermes-agent pattern; vitest/tsc --noEmit/eslint/oxfmt all clean locally before opening.
• ollama/ollama: docs: add protoAgent to Terminal & CLI integrations ollama/ollama#17738 — one line under ### Terminal & CLI, appended after the section's last entry (also not alphabetical — insertion-order convention).

Linked back on #1833 / #1834 (already closed — commented with the PR links there for traceability). Will close those + check the boxes here once each PR merges; the HF one is estimated ≈1–2 weeks review per the issue.

Not done: the "nice-to-have" GGUF/MLX quant-card section — deferred, no tags: conversational GGUF with a tool-calling chat_template picked yet.

[mabry1985]

Triage: 🔴 Reconsider — V2/LOE1, Risk Low. Explicitly deferred until more marketing + demos. Consider closing with a revisit label when marketing push is ready.