feat(standardcompute): add Standard Compute provider (flat-rate OpenAI-compatible router)

[henrikdukefoss]

Adds Standard Compute, a flat-rate unlimited LLM gateway with an OpenAI-compatible API.

• Provider: @ai-sdk/openai-compatible, base https://api.stdcmpt.com/v1, auth via STANDARDCOMPUTE_API_KEY
• Model: single standardcompute auto-route id (analogous to LLM Gateway's auto) — routes each request to a suitable frontier model
• Cost: per-token cost is $0 — pricing is subscription tiers ($9–$399/mo) that differ by execution speed, not usage caps (same convention as other plan-included $0 entries)
• Capabilities: tool_call and structured_output verified against the live endpoint on 2026-07-17; conservative on the rest (attachment/reasoning false, text-only modalities, 128k context floor)

Disclosure: I run Standard Compute. Happy to adjust fields to maintainer preference.

[github-actions]

Action items

• [high] [violation] providers/standardcompute/logo.svg - Check: New providers must ship a compliant logo. Why: AGENTS.md makes a logo a hard merge blocker for every new provider, and the diff only adds provider.toml and models/standardcompute.toml — no logo.svg. The directly analogous providers/llmgateway/ (the "LLM Gateway auto" router this PR mirrors) ships one. Action: Add providers/standardcompute/logo.svg using SVG format, currentColor for all fills/strokes, no fixed size or hardcoded colors, and preferably a square viewBox (e.g. 0 0 24 24).
• [low] [possible mistake] providers/standardcompute/models/standardcompute.toml:4 - Check: Material factual claims (release date, last-updated, token limits) should be supported by a cited source. Why: The PR body states only tool_call/structured_output were verified against the live endpoint on 2026-07-17 and describes 128k as a "floor," yet the file asserts release_date = "2026-03-01", last_updated = "2026-07-17", and a specific output = 16_384. The cited integrations page does not establish a launch date or an output-token cap, so these values cannot be reviewed from the included evidence. Action: Add a first-party citation (pricing/docs/API reference) supporting the release date and output limit, or document in the leading comment block that these are conservative placeholders and why.

[henrikdukefoss]

Both action items addressed:

• [high] logo: added providers/standardcompute/logo.svg — single path, fill="currentColor", square viewBox, no fixed dimensions or hardcoded colors.
• [low] claim provenance: the model TOML's leading comment now documents what was verified against the live endpoint (tool_call, structured_output, 2026-07-17), that release_date is the operator-stated public launch month, and that limit.context/limit.output are conservative floor placeholders per the auto-router convention — the router fans out to frontier backends whose true limits vary per request. attachment/reasoning are marked false pending verification rather than as claims of absence.

Happy to adjust further.

[github-actions]

I have gathered enough context to evaluate the PR. The proposed files are a new standardcompute provider (provider.toml, logo.svg, and a single standardcompute auto-route model). I verified the schema, family registry, logo guidelines, comment-placement rules, the base_model/reasoning_options requirements, and compared against existing auto-router providers (llmgateway/auto, orcarouter/auto, morph/auto).

Action items

• [low] [possible mistake] providers/standardcompute/provider.toml:6 - Check: Direct pricing citation for the subscription-based $0 per-token cost claim. Why: The only cited source is https://standardcompute.com/integrations, which is an integrations/docs URL, not a pricing page. The PR body and the leading comment assert subscription tiers ($9–$399/mo) as the justification for cost.input = 0 / cost.output = 0, but that material factual claim cannot be reviewed from the cited URL alone, and a bare URL comment does not state what it supports. The $0 value itself is consistent with other plan-included auto-router entries (e.g. llmgateway/auto, orcarouter/auto), so this is not a blocker, but the subscription-pricing rationale needs a direct first-party pricing reference. Action: Add a direct Standard Compute pricing/subscription page URL (in the PR body and/or the provider.toml leading comment block) that substantiates the flat-rate subscription model and $0 per-token pricing.

[henrikdukefoss]

Addressed: the provider.toml leading comment now cites the direct pricing page (https://standardcompute.com/pricing, accessed 2026-07-18) substantiating the subscription tiers ($9–$399/mo by execution speed) behind the $0 per-token cost, alongside the integrations doc.

[github-actions]

No actionable findings.