feat: add Atlas Cloud provider

[binyangzhu000-sudo]

Summary

• add Atlas Cloud as a built-in OpenAI-compatible provider alias
• discover ATLASCLOUD_API_KEY for automatic model selection and credential forwarding
• document inline and named-model configuration
• include the provider in navigation, the built-in alias table, and the generated schema description

Testing

• go test ./pkg/model/provider (focused alias and end-to-end provider tests)
• go test ./pkg/config ./pkg/runtime ./pkg/creator ./pkg/environment (focused provider/config tests)
• go test ./pkg/config ./pkg/model/provider ./pkg/model/provider/openai
• go build ./...
• go run ./lint .
• golangci-lint v2.12.2 run
• go mod tidy --diff
• CI=true ./scripts/build.sh
• live Atlas Cloud Chat Completions smoke test (HTTP 200)
• live GET https://api.atlascloud.ai/v1/models on 2026-08-17: HTTP 200, 136 models, and the default qwen/qwen3.8-max is present

A full go test ./... run passed all emitted packages except pkg/teamloader's DMR examples, which require a local Docker Model Runner at 127.0.0.1:12434; it was not running in this environment.

[aheritier]

👋 Some commits in this PR are not signed and verified by GitHub. Please sign your commits with a GPG or SSH key registered in your GitHub account, then force-push.

Commits that are not verified: 24eb110

See GitHub's guide on signing commits for setup instructions. I've added status/needs-signed-commits; it will be removed automatically once every commit in this PR carries a valid GitHub-verified signature.

[Sayt-0]

Adding Atlas Cloud as a built-in alias follows the established pattern and the Go changes are complete and well tested. Three integration points are missing and the default model needs verification before merge.

#	Finding	File	Blocking
1	New docs page absent from the sidebar	docs/data/nav.yml	yes
2	Built-in alias table not updated	docs/concepts/models/index.md	yes
3	Provider list in schema description not updated	agent-schema.json	no
4	Default model qwen/qwen3.8-max unverifiable	pkg/config/auto.go (see inline)	yes
5	Front matter weight breaks alphabetical convention	docs/providers/atlascloud/index.md (see inline)	no

Ready-to-paste fixes for the files outside the diff:

1. docs/data/nav.yml, between the Anthropic and AWS Bedrock entries:

    - title: Atlas Cloud
      url: /providers/atlascloud/

2. docs/concepts/models/index.md, built-in alias table, after the OpenRouter row (matches the placement used in docs/providers/overview/index.md; adjust the description column if needed):

| Atlas Cloud         | `atlascloud`     | Qwen and other open models           | `ATLASCLOUD_API_KEY`                |

3. agent-schema.json, provider description, add atlascloud after openrouter in the alias enumeration:

... any built-in alias (requesty, openrouter, atlascloud, azure, xai, ...)

Process notes:

• The status/needs-signed-commits label is set; the commit must be signed before merge.
• The branch is 11 commits behind main. A rebase also picks up 95f3d1a, which fixes the pre-existing TestParseExamples gopher failures in pkg/config, unrelated to this PR but noisy in local runs.

[Sayt-0]

Provider pages use alphabetical-by-title weights in increments of 10 (Anthropic 20, AWS Bedrock 30, Baseten 40, ...). "Atlas Cloud" sorts between Anthropic and AWS Bedrock, so 235 places the page between OpenRouter and OVHcloud in weight-ordered listings.

Suggested change

	weight: 235
	weight: 25

[Sayt-0]

Can the existence of qwen/qwen3.8-max in the Atlas Cloud catalog be confirmed (catalog link or output of the live smoke test mentioned in the PR description)? Two concerns:

• This ID is the auto-selection default: if it is wrong, every user with only ATLASCLOUD_API_KEY set gets a broken first run.
• Neither an atlascloud provider nor any qwen3.8-max entry exists in pkg/modelsdev/snapshot.json, so capability lookup falls back to text-only with a warning. A "-max" model is also unusual for an open-model host (the -max line is API-only, not open weights), which is worth double-checking against the openModelHostProviders classification.

If the ID changes, the two examples in docs/providers/atlascloud/index.md (atlascloud/qwen/qwen3.8-max and model: qwen/qwen3.8-max) and the assertions in pkg/config/auto_test.go and pkg/model/provider/openai_alias_providers_test.go must be updated in sync.

[Sayt-0]

Hello @binyangzhu000-sudo looks good now but one more thing before approve : your commits must be signed. Don't hesitate to ping me when it's done !

[aheritier]

👋 This PR has merge conflicts with the base branch. Please rebase or merge the latest base branch and resolve them. I've moved it to draft and added status/needs-rebase; it'll be picked back up automatically once the conflicts are cleared.

[binyangzhu000-sudo]

@aheritier Conflicts cleared — merged current main (597 commits).

Both conflicts were in the same place, where you refreshed default models next to the Atlas Cloud entry. I took your values for the two you changed and kept ours:

"openrouter":     "meta-llama/llama-4-maverick",   // yours
"atlascloud":     "qwen/qwen3.8-max",
"baseten":        "deepseek-ai/DeepSeek-V4-Pro",   // yours

with the matching assertions in auto_test.go. I re-checked qwen/qwen3.8-max against the Atlas catalogue rather than assuming it was still current — it is present, and it is the newest max model listed.

The merge also brought in TestDefaultModelsExistInModelsDev (the regression test for #4133), which requires every DefaultModels entry to resolve in the models.dev catalog. Atlas Cloud isn't listed there — its model ids come from its own /api/v1/models endpoint — so I registered it in modelsDevAbsentProviders, the escape hatch that test's own doc comment describes for exactly this case, rather than weakening the assertion.

Verification on the merged branch:

• go build ./... → clean
• go test ./pkg/config/... ./pkg/model/provider/... → all packages pass except TestURLSource_Read_RejectsLocalAddresses, which fails here because this machine can't route to 192.168.1.1 / 169.254.169.254 / 10.0.0.1 and the fetch times out instead of being rejected as a non-public address. I confirmed it fails identically in a clean upstream/main worktree, so it's environmental, not from this branch.
• TestDefaultModelsExistInModelsDev → passes.

@Sayt-0 — on the signing requirement: that's the one thing I can't do from here, since it needs the author's own signing key. I've flagged it to the account owner and it will be handled separately; I didn't want the rebase to keep blocking in the meantime. Note the merge commit above is also unsigned, so it will need to be included when the branch is re-signed and force-pushed.