Skip to content

feat(ai): add MiniMax provider with Flash-first model choices - #331

Open
octo-patch wants to merge 6 commits into
CoreBunch:mainfrom
octo-patch:octo/20260803-provider-add-recvqkMZVZFgyJ
Open

octo-patch wants to merge 6 commits into
CoreBunch:mainfrom
octo-patch:octo/20260803-provider-add-recvqkMZVZFgyJ

Conversation

@octo-patch

@octo-patch octo-patch commented Aug 3, 2026 •

Copy link
Copy Markdown

Reason: Instatic needs a first-class MiniMax connection with current ordinary model choices and protocol-correct tool continuation. New MiniMax credentials should prefer MiniMax-M3.1-Flash, then MiniMax-M3, while existing saved model IDs remain usable.

Summary

  • Add MiniMax provider connections for global and China OpenAI-compatible and Anthropic-compatible endpoints, requiring an API key and using each region's live model catalogue.
  • Offer Flash first and M3 second for direct connections. Update context/capability metadata and standard short-context prices. OpenRouter offers the verified minimax/minimax-m3.1-flash:free route before minimax/minimax-m3, retaining gateway prices/context and saved legacy model lookup. Other provider and media entries are preserved.
  • Preserve MiniMax reasoning during tool continuation: Chat Completions replays reasoning_content; Messages replays thinking blocks and signatures in their original order. Shared adapters enable this only for the direct MiniMax provider.
  • Keep protocol defaults by omitting thinking/effort overrides: Flash defaults to adaptive/max; M3 defaults to adaptive on Chat Completions and disabled on Messages. Instatic does not expose explicit thinking/effort controls. Existing persisted histories are unchanged; they do not store reasoning blocks.
  • Update the provider label, connection test, and user documentation. Direct static prices describe the standard short-context tier; the existing pricing shape does not represent M3 long-context or priority multipliers.

Sources: OpenAI-compatible API, Messages API, pricing, and live OpenRouter catalogue, inspected 2026-10-20.

Verification

  • bun run build — deferred under the shared runner's disk-resource pause; fewer than 110 GiB available.
  • bun test — full suite not run. Focused HTTP-boundary/regression tests could not load missing happy-dom. A small dependency setup at repository lockfile versions waited on the shared light-install lock; the initial 180-second wait expired and a longer wait was canceled for handoff after over 20 minutes without install output. Focused execution needs a retry.
  • bun run lint — full dependency setup is resource-paused.
  • Docker/deployment check, if relevant — not applicable.
  • Changed TypeScript syntax transpilation and git diff --check pass. Bun 1.3.14 is available locally (manifest supports >=1.2.4; CI pins 1.3.11).
  • Added offline HTTP-boundary tests for four regional/protocol endpoints, discovery-to-request default selection, saved/custom IDs, omitted thinking/effort/service-tier fields, tools, reasoning continuation, prior text history, images, and gateway catalogue policy. No paid inference was performed.
  • Exact head dbda218b8baaa36e5422127cd055b49b286e3c65: both Socket checks passed. Repository CI is action_required awaiting maintainer approval; its lint/build/test jobs have not run.

Checklist

  • Tests cover behavior changes.
  • Docs were updated when behavior, config, deployment, or public surfaces changed.
  • No compatibility shim was added for old pre-release behavior.
  • No secrets, local databases, uploads, or generated artifacts are included.

@octo-patch

Copy link
Copy Markdown
Author

Updated this PR with MiniMax global/China endpoint presets for both supported wire protocols, required API-key handling, current M3/M2.7 metadata, and protocol-specific model discovery and streaming coverage. Focused tests, TypeScript checking, ESLint, and the production build pass; the two unrelated full-suite failures reproduce on the pre-change PR head.

@octo-patch octo-patch changed the title feat(ai): add MiniMax provider feat(ai): add MiniMax provider with Flash-first model choices Oct 20, 2026

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