feat(providers): add MiniMax provider

[octo-patch]

Summary

Adds MiniMax as a first-class Bifrost provider with current model, endpoint, schema, UI, and documentation coverage.

Changes

• Registers MiniMax across core construction, configuration schemas, migration handling, UI metadata, and API documentation.
• Supports model listing, chat completions, streaming chat completions, and the native Responses API.
• Honors context path overrides for streaming requests and forwards MiniMax-specific request parameters.
• Preserves MiniMax image and video content fields through the shared chat schema and generated OpenAPI definitions.
• Covers MiniMax-M3 and MiniMax-M2.7, including multimodal input, thinking controls, context windows, regional endpoints, and tiered pricing.
• Returns an unsupported-operation error for classic text completions because the upstream API does not expose that route.
• Adds a local SSE regression test for path overrides, request parameter forwarding, and multimodal content forwarding.

Type of change

• Bug fix
• Feature
• Refactor
• Documentation
• Chore/CI

Affected areas

• Core (Go)
• Transports (HTTP)
• Providers/Integrations
• Plugins
• UI (React)
• Docs

How to test

cd core
env MINIMAX_API_KEY= GOWORK=off go test -count=1 ./providers/minimax
env MINIMAX_API_KEY= GOWORK=off go vet ./providers/minimax
env MINIMAX_API_KEY= GOWORK=off go test -run '^$' .

cd ..
env MINIMAX_API_KEY= PATH="$(go env GOPATH)/bin:$PATH" make test-core PROVIDER=minimax PATTERN=ContextPath

cd docs/openapi
python bundle.py

cd ../../ui
npm run build-enterprise
npm run typecheck

OpenAPI generation, Helm schema validation, the UI enterprise build, formatting, typecheck, and documentation parsing pass on the final tree. The MiniMax regression test, provider vet, and core compile check pass after locally removing the current dev branch's stale rawToolSearch assignment; the final tree inherits that pre-existing compile error and intentionally does not include the unrelated workaround. The live integration suite skips when no MiniMax API key is supplied.

Screenshots/Recordings

Not applicable. The UI change registers provider metadata and the MiniMax brand icon.

Breaking changes

• Yes
• No

Related issues

N/A

Security considerations

The API key remains stored through the existing secret variable mechanism and is sent only in the authorization header. No key values are logged or included in tests or documentation.

Checklist

• I read the repository contribution guidelines and followed them
• I added/updated tests where appropriate
• I updated documentation where needed
• I verified builds succeed (Go and UI)
• I verified the applicable CI checks locally

[CLAassistant]

All committers have signed the CLA.

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR adds MiniMax as a supported provider with OpenAI-compatible chat and Responses APIs, streaming support, provider registration, configuration, tests, documentation, media schemas, and UI metadata.

Changes

MiniMax provider integration

Layer / File(s)	Summary
Provider contracts and media schemas core/schemas/..., transports/..., docs/openapi/..., helm-charts/...	Registers MiniMax across provider allowlists and adds video input and pricing-related schema fields.
MiniMax provider implementation core/providers/minimax/...	Adds provider construction, model listing, chat and Responses APIs, streaming handlers, cached-content stubs, and unsupported-operation responses.
Provider construction and test configuration core/bifrost.go, core/internal/llmtests/..., core/providers/minimax/*_test.go	Wires provider creation and configures keys, network settings, expectations, comprehensive scenarios, custom paths, and forwarded parameters.
Documentation and UI metadata docs/providers/..., docs/docs.json, ui/lib/constants/..., core/changelog.md, transports/changelog.md	Adds MiniMax documentation, navigation and support-matrix entries, labels, placeholders, key requirements, icon metadata, and changelog entries.

Estimated code review effort: 4 (Complex) | ~60 minutes

Possibly related PRs

• maximhq/bifrost#4425: The provider uses the shared OpenAI request handlers affected by this auth-header refactor.
• maximhq/bifrost#4601: Both add provider registration and comprehensive-test integration.
• maximhq/bifrost#5033: Both add pricing and additional-attributes fields to the OpenAPI schema.

Suggested reviewers: akshaydeo, danpiths, pratham-mishra04, roroghost17, r-droid101

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title is concise and accurately summarizes the main change: adding MiniMax provider support.
Description check	✅ Passed	The description follows the template well and includes summary, changes, test steps, impact, security, and checklist items.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Confidence Score: 5/5

Safe to merge; the new provider follows established patterns with no changes to existing provider paths.

The MiniMax provider delegates entirely to the existing OpenAI-compatible helpers. Streaming uses the dedicated streaming client built by BuildStreamingClient. All unsupported operations return the standard unsupported-operation error. The shared schema additions (video_url, max_long_side_pixel) are purely additive with omitempty and do not alter serialization for any existing provider. No mutations to shared mutable state, no new goroutines, no fasthttp acquire/release concerns beyond what the existing helpers already manage.

No files require special attention; the most cross-cutting change is core/schemas/chatcompletions.go (shared schema extensions), which is backward-compatible.

Important Files Changed

Filename	Overview
core/providers/minimax/minimax.go	New MiniMax provider using OpenAI-compatible helpers; correctly uses BuildStreamingClient for streaming paths, BifrostContextKeyPassthroughExtraParams for extra params, bearer auth, and unsupported-operation errors for unimplemented operations.
core/providers/minimax/minimax_unit_test.go	Local SSE regression test verifying context path override, extra-params forwarding, and multimodal (image + video) content block passthrough; test provider reuses the same fasthttp.Client for both client and streamingClient but this doesn't affect correctness for the scenarios covered.
core/schemas/chatcompletions.go	Adds ChatContentBlockTypeVideo constant, ChatInputVideo struct, and MaxLongSidePixel field to ChatInputImage; all additions use omitempty so they are backward-compatible for all providers that don't populate these fields.
core/schemas/bifrost.go	Adds Minimax ModelProvider constant and adds it to StandardProviders (correctly omitted from SupportedBaseProviders, which is reserved for custom-provider base types).
transports/config.schema.json	Adds minimax as a recognized provider key in all three relevant locations (provider definitions, provider enum, and key provider enum), consistent with other providers.
ui/lib/constants/icons.tsx	Adds inline SVG MiniMax brand icon with a linear gradient; uses a fixed gradient ID (minimax-icon-gradient) consistent with the existing pattern for other icons in the same file.
docs/openapi/schemas/inference/chat.yaml	Adds video_url to ChatContentBlock enum and ChatInputVideo schema; extends ChatInputImage.detail enum with the default value used by MiniMax and adds max_long_side_pixel field.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant Caller
    participant Bifrost
    participant MinimaxProvider
    participant OpenAIHelper
    participant MiniMaxAPI

    Caller->>Bifrost: ChatCompletion / ChatCompletionStream
    Bifrost->>MinimaxProvider: route to provider
    MinimaxProvider->>MinimaxProvider: SetValue(PassthroughExtraParams, true)
    alt Unary
        MinimaxProvider->>OpenAIHelper: HandleOpenAIChatCompletionRequest(client, ...)
        OpenAIHelper->>MiniMaxAPI: POST /v1/chat/completions
        MiniMaxAPI-->>OpenAIHelper: JSON response
        OpenAIHelper-->>MinimaxProvider: BifrostChatResponse
        MinimaxProvider-->>Bifrost: BifrostChatResponse
    else Streaming
        MinimaxProvider->>OpenAIHelper: HandleOpenAIChatCompletionStreaming(streamingClient, ...)
        OpenAIHelper->>MiniMaxAPI: POST /v1/chat/completions (SSE)
        MiniMaxAPI-->>OpenAIHelper: SSE chunks
        OpenAIHelper-->>MinimaxProvider: chan BifrostStreamChunk
        MinimaxProvider-->>Bifrost: chan BifrostStreamChunk
    end
    Bifrost-->>Caller: response / stream

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant Caller
    participant Bifrost
    participant MinimaxProvider
    participant OpenAIHelper
    participant MiniMaxAPI

    Caller->>Bifrost: ChatCompletion / ChatCompletionStream
    Bifrost->>MinimaxProvider: route to provider
    MinimaxProvider->>MinimaxProvider: SetValue(PassthroughExtraParams, true)
    alt Unary
        MinimaxProvider->>OpenAIHelper: HandleOpenAIChatCompletionRequest(client, ...)
        OpenAIHelper->>MiniMaxAPI: POST /v1/chat/completions
        MiniMaxAPI-->>OpenAIHelper: JSON response
        OpenAIHelper-->>MinimaxProvider: BifrostChatResponse
        MinimaxProvider-->>Bifrost: BifrostChatResponse
    else Streaming
        MinimaxProvider->>OpenAIHelper: HandleOpenAIChatCompletionStreaming(streamingClient, ...)
        OpenAIHelper->>MiniMaxAPI: POST /v1/chat/completions (SSE)
        MiniMaxAPI-->>OpenAIHelper: SSE chunks
        OpenAIHelper-->>MinimaxProvider: chan BifrostStreamChunk
        MinimaxProvider-->>Bifrost: chan BifrostStreamChunk
    end
    Bifrost-->>Caller: response / stream

Loading

_{Reviews (5): Last reviewed commit: "[fix]: preserve MiniMax multimodal conte..." | Re-trigger Greptile}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@core/providers/minimax/minimax.go`:
- Around line 106-125: Streaming completion methods are ignoring the request
path from context because they hardcode the OpenAI endpoint. Update
MinimaxProvider.TextCompletionStream to use GetPathFromContext for the URL path
instead of a fixed "/v1/completions", and apply the same change to the analogous
streaming methods in the other providers named in the review. Keep the existing
openai.HandleOpenAITextCompletionStreaming/openai.HandleOpenAIChatCompletionStreaming
delegation, but pass the context-derived path so BifrostContextKeyURLPath is
honored consistently for streaming and non-streaming calls.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 858b9431-87b9-4f1d-b403-c7fc82466d95

📥 Commits

Reviewing files that changed from the base of the PR and between f244691 and d36435e.

📒 Files selected for processing (9)

• core/bifrost.go
• core/internal/llmtests/account.go
• core/internal/llmtests/validation_presets.go
• core/providers/minimax/cachedcontents.go
• core/providers/minimax/minimax.go
• core/providers/minimax/minimax_test.go
• core/schemas/bifrost.go
• core/utils.go
• scripts/bifrost-migration-cli/model.go

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Check whether sibling OpenAI-compatible providers (e.g., cerebras) apply GetPathFromContext consistently between unary and streaming chat/text completion calls.
rg -n -B2 -A2 'BaseURL\s*\+\s*"?(providerUtils\.GetPathFromContext)?' core/providers/cerebras/cerebras.go core/providers/groq/groq.go core/providers/perplexity/perplexity.go 2>/dev/null
rg -n 'GetPathFromContext' core/providers/cerebras core/providers/groq core/providers/perplexity core/providers/minimax

Repository: maximhq/bifrost

Length of output: 7515

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Find streaming completion calls in OpenAI-compatible providers and compare them
# with their unary counterparts for GetPathFromContext usage.
rg -n -C 2 'provider\.streamingClient|HandleOpenAI.*Streaming|GetPathFromContext' core/providers/{minimax,cerebras,groq,perplexity,openrouter,ollama,parasail,nebius,xai,sgl} 2>/dev/null

# Focused scan for hardcoded streaming URLs that skip GetPathFromContext.
rg -n 'streamingClient.*"\s*\+|BaseURL\s*\+\s*"/v1/|BaseURL\s*\+\s*"/chat/' core/providers/{minimax,cerebras,groq,perplexity,openrouter,ollama,parasail,nebius,xai,sgl} 2>/dev/null

Repository: maximhq/bifrost

Length of output: 29685

---

Streaming paths should honor GetPathFromContext
TextCompletionStream/ChatCompletionStream hardcode the OpenAI path, so BifrostContextKeyURLPath is ignored for streaming requests while still working for non-streaming calls. The same pattern also shows up in Cerebras, Groq, OpenRouter, Parasail, Nebius, Ollama, xAI, SGL, and Perplexity, so the fix should be applied wherever streaming delegates to the OpenAI helpers.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@core/providers/minimax/minimax.go` around lines 106 - 125, Streaming
completion methods are ignoring the request path from context because they
hardcode the OpenAI endpoint. Update MinimaxProvider.TextCompletionStream to use
GetPathFromContext for the URL path instead of a fixed "/v1/completions", and
apply the same change to the analogous streaming methods in the other providers
named in the review. Keep the existing
openai.HandleOpenAITextCompletionStreaming/openai.HandleOpenAIChatCompletionStreaming
delegation, but pass the context-derived path so BifrostContextKeyURLPath is
honored consistently for streaming and non-streaming calls.

[Pratham-Mishra04]

Hey @octo-patch this can be used as a custom provider right? any specific reason to add it as a first party provider?

[octo-patch]

Yes, its OpenAI-compatible surface can be configured as a custom provider. First-party support removes per-deployment custom configuration and provides a stable built-in identity and default endpoint, schema/UI/migration registration, documented regional setup, model listing, provider-specific parameter passthrough, and dedicated regression coverage.

[octo-patch]

This PR has been replaced by #5184. The replacement is a single clean commit based on the latest dev branch and carries the provider implementation, tests, configuration, UI, and documentation changes.