Skip to content

feat(ai): add Atlas Cloud provider - #758

Open
binyangzhu000-sudo wants to merge 1 commit into
markrahq:v2from
binyangzhu000-sudo:add-atlascloud-provider
Open

binyangzhu000-sudo wants to merge 1 commit into
markrahq:v2from
binyangzhu000-sudo:add-atlascloud-provider

Conversation

@binyangzhu000-sudo

Copy link
Copy Markdown

Summary

Adds Atlas Cloud (https://api.atlascloud.ai/v1, an OpenAI-compatible gateway) as a disabled built-in provider, wired the same way as the existing OrcaRouter entry (#718): a catalog template on the existing OpenAI-compatible transport, model discovery enabled, and a migration that appends it to already-saved settings.

Disclosure: I work at Atlas Cloud.

Files:

  • packages/providers/src/catalog.ts: new atlascloud template (disabled, default deepseek-ai/deepseek-v4-flash, 5 models)
  • packages/providers/src/provider-capabilities.ts: atlascloud: { modelList: true }
  • packages/providers/src/thinking-format.ts: reasoning_effort: "high" when thinking is on, omitted when off
  • packages/providers/src/settings.ts: migration loop now appends both orcarouter and atlascloud to existing saved settings
  • README.md, README.zh-CN.md, .github/ISSUE_TEMPLATE/ai-provider.yml: provider lists
  • Tests: settings, provider-capabilities, requests, thinking-format, and the app ai-settings provider-id list

Decisions:

  • Reasoning follows the OrcaRouter precedent: the gateway fronts many upstreams and only high/xhigh effort is accepted by all of them, so nothing is sent when thinking is off instead of guessing per-model disable fields.
  • No logo asset added; the badge falls back to the existing text avatar.

Test plan

  • pnpm test in packages/providers: 53 passed
  • pnpm test in packages/ai: 264 passed
  • pnpm test in packages/app: 1644 passed (141 files)
  • pnpm typecheck:test in packages/providers: clean
  • Live check through the repo's own code (temporary vitest file, not committed): fetchAiProviderModels against the real gateway returned 119 models; getChatAdapterForProvider(...).buildRequest for deepseek-ai/deepseek-v4-flash with thinking on produced POST /v1/chat/completions with reasoning_effort: "high" and got HTTP 200 with content and reasoning_content
  • Not run: pnpm build and the desktop app UI (no manual check of the settings screen); typecheck:test for ai/app

🤖 Generated with Claude Code

Add a disabled built-in Atlas Cloud provider (https://api.atlascloud.ai/v1)
that reuses the existing OpenAI-compatible transport, mirroring how the
OrcaRouter gateway entry is wired.

- catalog template with a small default model list
- model discovery enabled via GET /v1/models
- reasoning maps to reasoning_effort "high" when thinking is on and is
  omitted when off, since only high/xhigh are accepted across upstreams
- existing saved settings get the entry appended without touching saved
  providers or selections
- README (en/zh-CN) and issue template provider lists updated

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Signed-off-by: binyangzhu000-sudo <224954946+binyangzhu000-sudo@users.noreply.github.com>
@vercel

vercel Bot commented Oct 14, 2026

Copy link
Copy Markdown

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

A member of the Team first needs to authorize it.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant