[binyangzhu000-sudo]

Summary

Adds Atlas Cloud as a built-in provider card in the extended catalog, using the existing OpenAI-compatible adapter. It follows the same pattern and file set as the Cheaper Inference card (#3268).

• Card id atlascloud, category router, disabled by default.
• Base URL https://api.atlascloud.ai/v1.
• Default model qwen/qwen3.7-plus (1000000 tokens, vision on). Suggestions are ids from the public GET https://api.atlascloud.ai/v1/models list: qwen/qwen3.7-plus, deepseek-ai/deepseek-v4-flash, deepseek-ai/deepseek-v3.2, moonshotai/kimi-k2.6.
• Chrome and Firefox catalogs stay identical.
• No pricing metadata is added.

I work at Atlas Cloud.

Files

• src/{chrome,firefox}/src/providers/provider-catalog.js: one new entry, placed after alibaba-coding-plan-cn.
• src/{chrome,firefox}/icons/providers/atlascloud.svg: the Atlas Cloud mark from Lobe Icons (@lobehub/icons-static-svg, MIT), byte-identical to the upstream file.
• src/{chrome,firefox}/icons/providers/SOURCES.md: atlascloud added to the existing Lobe Icons row.
• test/run.js: id list, counts (84 extra cards, 119/118 defaults) and a defaults check.
• docs/providers-and-models.md: id table, counts and a short setup section.
• README.md: card count 118/117.

Verification

Run locally on a checkout of current main (a104a9a) plus this commit, Node v24.14.0, after npm ci --ignore-scripts:

• WEBBRAIN_TEST_FILTER=provider node test/run.js: 119 passed, 1 failed. The failure ("the Providers page LLM tab has a search box beside provider filters") fails the same way on unmodified main. The extended catalog test passes.
• Full node test/run.js: 2542 passed, 4 failed. The same 4 fail on unmodified main (identical count and names): "new locales are registered in extension and web language dropdowns", "sidepanel suppresses streamed raw tool-call text before rendering tool steps", "the Providers page LLM tab has a search box beside provider filters", "WebGPU worker follows local text-generation and WebBrain VL vision contracts".
• node test/provider-catalog-integrations.mjs and node test/provider-model-limits.mjs: pass.
• node --check on both catalogs and test/run.js: pass. git diff --check: clean.
• npm run build:web leaves no diff in tracked files.

Live calls through WebBrain's own OpenAICompatibleProvider (Chrome and Firefox copies, with the config taken from ADDITIONAL_PROVIDER_DEFAULTS.atlascloud), against https://api.atlascloud.ai/v1 with qwen/qwen3.7-plus:

• testConnection(): {ok: true} on both copies; an invalid key returns atlascloud error 401 on both.
• chat(): returned OK.
• chatStream(): streamed 1, 2, 3, 4, 5.
• Tool call: a forced-use get_weather tool returned a well-formed tool call with {"city": "Paris"}.
• Vision: a 128x128 solid red PNG was identified as Red.

This was run as a Node script that imports the catalog entry and the provider class, not inside a browser. I did not load the unpacked extension in Chrome or Firefox, so the Settings UI rendering of the card and the icon are untested by me (the catalog test checks the SVG exists and is well-formed).

Not done / caveats

• qwen/qwen3.7-plus always emits reasoning tokens (about 120 for a trivial prompt), so a very small max_tokens can return an empty content. Normal agent budgets are not affected, and Test Connection passes.
• The gateway's /v1/models supported_features field does not list reasoning/tools for this model, so the card's capability flags come from the live calls above, not from that field.
• deepseek-ai/deepseek-v4-flash and deepseek-ai/deepseek-v3.2 (suggestions) are text-only; the docs section says so. Switching the model does not change the card-level vision flag.
• I did not run the suite with the large benchmark result directories (test/llm/results*, test/llm/analysis, test/vision, test/llm-tiny*); my checkout excluded them.
• The CONTRIBUTING and GOVERNANCE text suggests proposing new LLM providers in an issue first; recent provider cards (feat(providers): add Cheaper Inference #3268, providers: add API Route card #3277, feat(providers): add Opper #3278) were merged without one, so I did not open an issue.
• Localized docs (docs/fr, docs/zh-CN) and the localized READMEs are untouched, as in feat(providers): add Cheaper Inference #3268.

🤖 Generated with Claude Code

[vercel]

@binyangzhu000-sudo is attempting to deploy a commit to the esokullu's projects Team on Vercel.

A member of the Team first needs to authorize it.