feat(rodiumai): add sync provider with base_model and reasoning_options#2159
feat(rodiumai): add sync provider with base_model and reasoning_options#2159Docteur-Parfait wants to merge 11 commits into
Conversation
RodiumAi exposes an **OpenAI-compatible** API at `https://api.rodiumai.io/v1`. Model IDs in this PR match the slugs returned by `GET /v1/models` on the live gateway.
Add RodiumAi as an OpenAI-compatible gateway (api.rodiumai.io) with smart routing tiers (auto, fast, basic, pro, max) and proxied upstream models. Register the rodium-smart model family and add required release_date and last_updated fields so bun validate passes in CI.
Add RodiumAi to bun models:sync from GET https://api.rodiumai.io/v1/models. Sync coding/chat models only (tools + text output; excludes embeddings, image, video). Vendor slugs factor canonical metadata via base_model; every model declares reasoning_options. Includes five smart-routing profiles and rodiumai-sync unit tests.
| last_updated = "2026-06-12" | ||
| reasoning = false | ||
| structured_output = true | ||
| reasoning_options = [] |
There was a problem hiding this comment.
it doesnt support reasoning budgets?
| reasoning = false | ||
| temperature = true | ||
| structured_output = true | ||
| reasoning_options = [] |
There was a problem hiding this comment.
no reasoning efforts?
|
a lot of the reasoning options seem incorrect here im skeptical |
|
Good catch — the first pass incorrectly defaulted Anthropic passthrough models to reasoning_options = [] because RodiumAi's supports_reasoning flag is false for adaptive-thinking models (Opus 4.7/4.8) even though they expose effort controls. Updated to mirror providers/anthropic/models/:
|
|
okay but can u check all ur models plz. like i said i doubt this is correct |
|
Alright I will check all of that and submit a new commit |
…provider models now use base_model + OpenRouter-aligned overrides (cost, reasoning_options, limits from live API). Re-run bun rodiumai:sync to refresh. Co-authored-by: Cursor <cursoragent@cursor.com>
|
Removed synthetic smart profiles (auto/basic/fast/pro/max). RodiumAI provider models now use base_model + OpenRouter-aligned overrides (cost, reasoning_options, limits from live API). Re-run bun rodiumai:sync to refresh. |
|
AUTOMATED REVIEW: The model definitions pass the |
3 participants
Summary
providers/rodiumai/, OpenAI-compatible API athttps://api.rodiumai.io/v1).bun models:sync rodiumai(packages/core/src/sync/providers/rodiumai.ts) — syncs coding/chat models fromGET /v1/models(tools + text output; excludes embeddings, image, video) plus five smart-routing profiles (auto,basic,fast,pro,max).base_modelwhere canonicalmodels/metadata exists; every model declaresreasoning_options(Anthropic effort/toggle defaults or[]).Provider details
RODIUMAI_API_KEY@ai-sdk/openai-compatibleaggregators(hourly automation via existing workflow)Test plan
bun rodiumai:syncbun validatebun test packages/core/test/rodiumai-sync.test.tsbun models:sync rodiumai --dry-run(idempotent: 55 unchanged)opencode --model rodiumai/auto,rodiumai/anthropic/claude-sonnet-4-6Links