Add Ace Data Cloud as a named OpenAI-compatible provider

[acedatacloud-dev]

Ace Data Cloud would like to contribute and maintain a named provider in Cline. We are submitting this Feature Request before implementation, as required by the contributor guidelines.

Today, users must select OpenAI Compatible and manually enter the base URL and model. The proposed built-in would expose Ace Data Cloud, a credentials-page link, and a model picker, while reusing Cline's existing OpenAI-compatible transport.

Configuration: provider ID acedatacloud; base URL https://api.acedata.cloud/openai; Bearer API key from ACEDATACLOUD_API_KEY or the saved password field; initially gpt-4.1 on Chat Completions. Credentials: https://platform.acedata.cloud/console/credentials ; API guide: https://platform.acedata.cloud/documents/openai-chat-completions .

On 2026-10-06 we verified authenticated model discovery and a two-request streamed tool loop against the production API: read_fixture({"path":"fixture.py"}), the matching tool result, and a streamed correction of return a - b to return a + b. Both responses were HTTP 200 with usage and terminal [DONE]; token counts were 91/19 and 133/15 input/output. This verifies the wire protocol for that model; it does not yet claim a completed Cline UI/agent test or support for all catalog models, vision, cache behavior, Responses, or Messages.

The implementation would cover SDK builtins/IDs, models.dev catalog mapping, VS Code known-provider and generic settings registration, focused configuration/streaming tests, and the provider setup docs. We reviewed the shared SDK model layer and the SDK-only ai& addition plus VS Code settings follow-up (#14439, #14611), so the proposed scope includes the actual configuration experience. Ace Data Cloud's proposed models.dev catalog is submitted in anomalyco/models.dev#8873. That proposed catalog is under review; model-control questions, maintainer acceptance, and publication are still pending. Catalog availability and Cline's generated snapshot must be verified before claiming a populated model picker. Public per-model cost expressions and package rates are available from https://platform.acedata.cloud/api/v1/models/catalog/?modality=chat . Price metadata will use the documented package conversion and will disclose that effective USD prices vary by package/account; other providers' prices will not be copied.

Would the maintainers accept this named-provider contribution and scope? If onboarding requires a separate provider review or partnership channel, please direct us to that process. This request is for provider discoverability and configuration; it does not request a default or recommended placement.

Brand asset

The official logo is this original PNG (512 × 512, RGBA, 60,479 bytes). Its SHA-256 is 3e2c6e2d1a61a549dae2374ec4444c77f4c2617c8f3772bdd4cd6e52bf20767e.

The preview uses a byte-identical copy pinned to a public commit because GitHub's image proxy could not fetch the canonical CDN URL during verification; the CDN link above remains the brand source.

Any logo shown in the eventual provider entry or setup material should retain this asset's proportions, gradient, transparency, and built-in whitespace. Please reuse the PNG where supported, rather than a tightly cropped or redrawn mark. The original has no outline; any icon container treatment should follow Cline's existing UI. This is supplied as branding for the proposal, not a claim that the provider has shipped.

Submitted by Ace Data Cloud Dev on behalf of Ace Data Cloud. This proposal and its validation were prepared with AI assistance; no claim of independent human code review is made.

[acedatacloud-dev]

Follow-up validation using the published Cline packages is now complete.

• @cline/agents@0.0.90 and @cline/llms@0.0.90, run with Bun 1.4.2, using the existing openai-compatible provider and https://api.acedata.cloud/openai.
• A real AgentRuntime called a local read_fixture tool, read an arithmetic bug, returned the tool result through the gateway, and streamed def add(a, b): return a + b.
• Result: completed, 2 iterations, exactly 2 HTTP 200 requests, usage included in both streams, 250 input / 34 output tokens, 5.554 seconds. Three local assertions against the generated function passed. The model output cap was 128 tokens per request.

This validates the released agent/gateway path through custom configuration. The release still has no named Ace Data Cloud entry, and product UI/distribution are not yet validated. The generic configuration reports totalCost: 0 because it lacks our pricing metadata; this does not mean the requests are free and reinforces the need for accurate per-model catalog prices.

No upstream feature implementation has started; we are awaiting the required maintainer approval for the scope in this proposal. Credentials were kept private. This additional validation was also AI-assisted.