[heydryft]

Summary

This PR adds Runcrate as a built-in aggregator preset. It is shaped like the Atlas Cloud and Cheaper Inference entries (#177, #184).

Runcrate serves open models (DeepSeek, GLM, Qwen, Kimi, gpt-oss) behind an OpenAI-compatible API. One API key works for OpenAI Chat, OpenAI Responses and Anthropic Messages.

The preset is not marked as recommended.

Changes

• presets.ts: new runcrate preset after cheaperinference, base https://api.runcrate.ai/v1. Three endpoints: OpenAI Chat, OpenAI Responses, and Anthropic (base https://api.runcrate.ai, the client adds /v1/messages). GET /v1/models works without a key, so the OpenAI Chat endpoint is marked modelListPublic. Suggested models are the full IDs from the model list: deepseek-ai/DeepSeek-V4.1-Flash, zai-org/GLM-5.3, moonshotai/Kimi-K2.6. The note says IDs carry the vendor prefix and are case sensitive, that the public list does not prove a key is valid, and that the same key works on all three protocols.
• baseurl.ts: Runcrate added to the /v1 row of the doc table.
• presets.test.ts: pinned the Anthropic URL row, and added runcrate to the modelListPublic set (six to seven).
• preset-brands.test.ts: lobehub has no Runcrate glyph, so it falls back to initials like Cheaper Inference. The count (50 to 51), the no-icon comment and the list are updated.

Keys: https://www.runcrate.ai/dashboard/api-keys?utm_source=NextCoWork&utm_medium=directory&utm_campaign=gpu_pricing
Docs: https://www.runcrate.ai/docs/models/overview?utm_source=NextCoWork&utm_medium=directory&utm_campaign=gpu_pricing

Testing

npx vitest run src/shared/domain src/renderer/src/settings/pages/model
# Test Files 52 passed (52), Tests 1393 passed (1393)
npm run typecheck:node   # passes
npm run typecheck:web    # passes
npx eslint src/shared/domain/presets.ts src/shared/domain/baseurl.ts src/shared/domain/__tests__/presets.test.ts src/renderer/src/settings/pages/model/__tests__/preset-brands.test.ts   # no findings

Verification is probed, with one difference from the usual probe: without a key every path under api.runcrate.ai returns 401, so the fake path comparison was done with a key. Results (2026-10-28):

• GET https://api.runcrate.ai/v1/models without a key: 200, model list returned.
• POST /v1/chat/completions, /v1/responses, /v1/messages without a key: 401.
• With a key, POST /v1/zzfake: 404.
• With a key, POST /v1/chat/completions on deepseek-ai/DeepSeek-V4.1-Flash: 200, reply "ok".
• With a key, POST /v1/responses on zai-org/GLM-5.3-Flash: 200, status: completed, output text "ok".
• With a key, POST /v1/messages (x-api-key, anthropic-version: 2023-06-01) on Qwen/Qwen3.8-Flash: 200, stop_reason: end_turn, text block "ok".

This PR was prepared with AI assistance: an AI coding agent wrote the change and ran the checks above.

Disclosure: I work on Runcrate.