Srihari Sriraman shared in

Public

Analyzed and summarized GitHub issue #994Public

Agent

Model

Claude Opus 4.5 (latest)

Started

Jan 29, 1:23 AM

Messages269

Downloads0

Summary

Closed, Completed

Understand
Reviewed issue and related comments.
Understand
Understand
Read related code files fully.
Understand
Understand
Checked external resources for context.
Understand
Verify
Confirmed issue auto-closed by commit.
Verify
Edit
Added summary comment to issue.
Edit

Analyze GitHub issue(s): https://github.com/badlogic/pi-mono/issues/994

For each issue:

Read the issue in full, including all comments and linked issues/PRs.
For bugs:
- Ignore any root cause analysis in the issue (likely wrong)
- Read all related code files in full (no truncation)
- Trace the code path and identify the actual root cause
- Propose a fix
For feature requests:
- Read all related code files in full (no truncation)
- Propose the most concise implementation approach
- List affected files and changes needed

Do NOT implement unless explicitly asked. Analyze and propose only.

Let me read the relevant files to understand the implementation scope:

Let me check what models.dev returns for Hugging Face:

Now let me check the documentation files to understand the format:

Analysis: Issue #994 - Add Hugging Face as a Provider

Type: Feature Request
Status: Open

Summary

The request is to add Hugging Face as a provider. The implementation is straightforward because:

Hugging Face Inference Router provides an OpenAI-compatible endpoint (https://router.huggingface.co/v1)
models.dev already tracks Hugging Face models with tool_call: true (16 models currently)
No new streaming implementation needed (reuses openai-completions)

models.dev Data Available

Key: huggingfaceAPI: https://router.huggingface.co/v1Env: HF_TOKENNPM: @ai-sdk/openai-compatible
Tool-capable models (16):- moonshotai/Kimi-K2-Instruct- moonshotai/Kimi-K2.5 (reasoning)- deepseek-ai/DeepSeek-R1-0528- deepseek-ai/DeepSeek-V3.2- Qwen/Qwen3-235B-A22B-Thinking-2507 (reasoning)- Qwen/Qwen3-Coder-480B-A35B-Instruct- MiniMaxAI/MiniMax-M2.1- zai-org/GLM-4.7- ... and more

Proposed Implementation

1. `packages/ai/src/types.ts`

Add "huggingface" to KnownProvider type union:

2. `packages/ai/src/env-api-keys.ts`

Add mapping in envMap:

3. `packages/ai/scripts/generate-models.ts`

Add Hugging Face model processing (similar to Groq/Cerebras pattern):

typescript

// Process Hugging Face modelsif (data.huggingface?.models) {	for (const [modelId, model] of Object.entries(data.huggingface.models)) {		const m = model as ModelsDevModel;		if (m.tool_call !== true) continue;
		models.push({			id: modelId,			name: m.name || modelId,			api: "openai-completions",			provider: "huggingface",			baseUrl: "https://router.huggingface.co/v1",			reasoning: m.reasoning === true,			input: m.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],			cost: {				input: m.cost?.input || 0,				output: m.cost?.output || 0,				cacheRead: m.cost?.cache_read || 0,				cacheWrite: m.cost?.cache_write || 0,			},			compat: {				supportsDeveloperRole: false,			},			contextWindow: m.limit?.context || 4096,			maxTokens: m.limit?.output || 4096,		});	}}

4. `packages/coding-agent/src/core/model-resolver.ts`

Add default model:

5. Tests

Add huggingface to provider lists in:

packages/ai/test/stream.test.ts
packages/ai/test/tokens.test.ts
packages/ai/test/abort.test.ts
packages/ai/test/empty.test.ts
packages/ai/test/context-overflow.test.ts
packages/ai/test/unicode-surrogate.test.ts
packages/ai/test/tool-call-without-result.test.ts
packages/ai/test/total-tokens.test.ts
packages/ai/test/cross-provider-handoff.test.ts (add at least one provider/model pair)

6. Documentation

packages/ai/README.md: Add to providers table and document HF_TOKEN env var.

packages/ai/CHANGELOG.md: Add under ## [Unreleased] → ### Added:

Open Questions

Compat settings: The issue suggests supportsDeveloperRole: false. This is likely correct (standard OpenAI-compatible endpoints typically don't support the newer developer role), but should be verified with actual API testing.
Default model choice: deepseek-ai/DeepSeek-V3.2 or moonshotai/Kimi-K2.5 are reasonable defaults. Need to pick one that's reliably available.
Video input: Some HF models (like Kimi-K2.5) support video input according to models.dev. The current Model type only supports "text" | "image". This is out of scope for this issue.

/var/folders/49/l4171l3n219_xwq30fmyzr8w0000gn/T/TemporaryItems/NSIRD_screencaptureui_XJ3dKg/Screenshot\ 2026-01-29\ at\ 02.28.15.png seeing this on models.dev.

does the json returns for each model actually tell us what vercel ai sdk is being used for that model?

Good question. Let me check the actual JSON structure more closely:

Yes! The JSON has the SDK info at two levels:

1. Provider level (applies to all models):

2. Model level (optional override):

For Hugging Face, all models inherit the provider-level npm: "@ai-sdk/openai-compatible", confirming api: "openai-completions" is correct.

The SDK → API mapping used by generate-models.ts (see OpenCode processing):

npm package	pi-ai api
`@ai-sdk/openai`	`openai-responses`
`@ai-sdk/anthropic`	`anthropic-messages`
`@ai-sdk/google`	`google-generative-ai`
`@ai-sdk/openai-compatible` / null	`openai-completions`

Kimi-K2.5 is the model i'd use in the tests

check if all huggingface models are labaled with openai-compatible

All 16 models have npm: null, meaning they all inherit the provider-level @ai-sdk/openai-compatible. No per-model overrides needed.

This simplifies the implementation: all Hugging Face models use api: "openai-completions" with the same base URL (https://router.huggingface.co/v1).

Updated recommendation for default model and tests:

Default: moonshotai/Kimi-K2.5 (reasoning-capable, multimodal)
Test model: moonshotai/Kimi-K2.5

ok, please implement

I'll implement the Hugging Face provider. Let me start by reading the files I need to modify in full:

Now I'll implement the changes. Let me start with the core files:

Now let me update the test files. First, let me read them to find the right places to add Hugging Face tests:

Now let me regenerate the models and run type checking:

Let me verify the generated models include Hugging Face:

All checks pass. Here's a summary of the implementation:

Changes Made

packages/ai/src/types.ts

Added "huggingface" to KnownProvider type union

packages/ai/src/env-api-keys.ts

Added huggingface: "HF_TOKEN" to the env var mapping

packages/ai/scripts/generate-models.ts

Added processing for data.huggingface?.models
Uses api: "openai-completions" with baseUrl: "https://router.huggingface.co/v1"
Sets compat: { supportsDeveloperRole: false }

packages/ai/src/models.generated.ts (auto-generated)

Now includes 14 Hugging Face models

packages/coding-agent/src/core/model-resolver.ts

Added huggingface: "moonshotai/Kimi-K2.5" to defaultModelPerProvider

packages/ai/CHANGELOG.md

Added entry under ## [Unreleased] → ### Added

Test files updated

stream.test.ts - Added Hugging Face provider tests
tokens.test.ts - Added Hugging Face provider tests
empty.test.ts - Added Hugging Face provider tests
context-overflow.test.ts - Added Hugging Face provider tests
unicode-surrogate.test.ts - Added Hugging Face provider tests
tool-call-without-result.test.ts - Added Hugging Face provider tests
total-tokens.test.ts - Added Hugging Face provider tests
cross-provider-handoff.test.ts - Added Hugging Face model pair

All tests use moonshotai/Kimi-K2.5 as specified, and are gated by HF_TOKEN env var.

docs in @packages/coding-agent/docs/models.md and @packages/coding-agent/docs/providers.md up-to-date? @README.md up-to-date=

The docs need updating. providers.md and README.md are missing Hugging Face. Let me fix that:

Done. Updated documentation:

docs/providers.md - Added Hugging Face | HF_TOKEN to the API Keys table
README.md - Added Hugging Face to the API keys list

All checks pass.

../shittycodingagent.ai/src/index.html also may need an update

The website mentions specific providers. Let me add Hugging Face:

Done. Added "Hugging Face" to the providers list on the website.

commit that then run publish.sh to publish

commit and push i mean

I'll only commit the index.html change (the README change looks unrelated):

Done. Committed and published to shittycodingagent.ai.

commit and push in this repo now, ensure we have CHANGELOG.md entry as well, then close issue with info in my tone

Loading more messages…

Summary

Analysis: Issue #994 - Add Hugging Face as a Provider

Summary

models.dev Data Available

Proposed Implementation

1. packages/ai/src/types.ts

2. packages/ai/src/env-api-keys.ts

3. packages/ai/scripts/generate-models.ts

4. packages/coding-agent/src/core/model-resolver.ts

5. Tests

6. Documentation

Open Questions

Changes Made

packages/ai/src/types.ts

packages/ai/src/env-api-keys.ts

packages/ai/scripts/generate-models.ts

packages/ai/src/models.generated.ts (auto-generated)

packages/coding-agent/src/core/model-resolver.ts

packages/ai/CHANGELOG.md

Test files updated

Summary

Analysis: Issue #994 - Add Hugging Face as a Provider

Summary

models.dev Data Available

Proposed Implementation

1. packages/ai/src/types.ts

2. packages/ai/src/env-api-keys.ts

3. packages/ai/scripts/generate-models.ts

4. packages/coding-agent/src/core/model-resolver.ts

5. Tests

6. Documentation

Open Questions

Changes Made

packages/ai/src/types.ts

packages/ai/src/env-api-keys.ts

packages/ai/scripts/generate-models.ts

packages/ai/src/models.generated.ts (auto-generated)

packages/coding-agent/src/core/model-resolver.ts

packages/ai/CHANGELOG.md

Test files updated

1. `packages/ai/src/types.ts`

2. `packages/ai/src/env-api-keys.ts`

3. `packages/ai/scripts/generate-models.ts`

4. `packages/coding-agent/src/core/model-resolver.ts`

1. `packages/ai/src/types.ts`

2. `packages/ai/src/env-api-keys.ts`

3. `packages/ai/scripts/generate-models.ts`

4. `packages/coding-agent/src/core/model-resolver.ts`