feat: add MiniMax provider support

[Rutetid]

[Provider] Add MiniMax Support

Closes #1855
/claim #1855

Summary

This PR adds complete support for MiniMax AI

Changes

Backend Integration

Core Implementation:

• ✅ Adapter (backend/src/routes/proxy/adapterV2/minimax.ts)

  • OpenAI-compatible request/response adapter with MiniMax-specific extensions
  • Streaming support with SSE parsing
  • Token estimation using tiktoken (MiniMax streaming returns usage: null)
  • Reasoning content extraction from reasoning_details array
  • Tool call handling and formatting
  • TOON compression for tool results
  • Model optimization support
  • Custom error handling for MiniMax's unique error format

• ✅ Type Definitions (backend/src/types/llm-providers/minimax/ - 4 files)

  • Complete Zod schemas for API types (api.ts, messages.ts, tools.ts, index.ts)
  • Message, tool, and request/response schemas
  • Extended schema for reasoning_details thinking content
  • Stream chunk type allows role: "" (empty string) - MiniMax quirk
  • OpenAI-compatible format with MiniMax extensions

• ✅ Routes (backend/src/routes/proxy/routesv2/minimax.ts)

  • HTTP proxy endpoints with agent support
  • Chat completions (streaming and non-streaming)
  • Error handling and validation
  • Workaround for missing /v1/models endpoint - uses hardcoded model list

• ✅ Database Migration (backend/src/database/migrations/0131_add_minimax_token_prices.sql)

  • Token pricing for all MiniMax models
  • MiniMax-M2: $0.30 input / $1.20 output per million tokens
  • MiniMax-M2.1: $0.30 input / $1.20 output per million tokens
  • MiniMax-M2.1-lightning: $0.30 input / $2.40 output per million tokens

• ✅ Models Dev Client Integration (backend/src/clients/models-dev-client.ts)

  • Added minimax provider mapping for automatic price sync

Frontend Integration

• ✅ Interaction Handler (frontend/src/lib/llmProviders/minimax.ts)

  • Message parsing and rendering
  • Reasoning content display (thinking)
  • Tool call display
  • Token usage calculation
  • Stop reason handling

• ✅ UI Components

  • Model selector integration
  • API key form with MiniMax branding
  • Provider display name
  • Chat interface support

Key MiniMax Differences

1. No Models Endpoint

MiniMax doesn't provide a /v1/models endpoint. We use a hardcoded list:

• MiniMax-M2
• MiniMax-M2.1
• MiniMax-M2.1-lightning (fastest)

2. Streaming Usage Data

Problem: MiniMax streaming API returns "usage": null in all chunks (no token counts)
Solution: Implemented token estimation using tiktoken that:

• Counts input tokens from request messages
• Counts output tokens from accumulated text, reasoning, and tool calls
• Runs automatically at stream end to ensure interactions are logged

3. Reasoning Details

MiniMax supports extended thinking with reasoning_details array:

• Enable with extra_body: { reasoning_split: true }
• Returns thinking process separate from main content
• Adapter extracts reasoning deltas and accumulates full text

4. Stream Chunk Role Field

MiniMax sends "role": "" (empty string) in some stream chunks instead of "role": "assistant". Type schema updated to allow both.

Streaming Support

• ✅ Non-streaming responses: Fully supported with native usage data
• ✅ Streaming responses: Fully supported with SSE + token estimation

Feature Completeness

LLM Proxy ✅

Feature	Status	Notes
Tool invocation	✅ Supported	Full tool call format conversion
Tool persistence	✅ Supported	Tool calls maintained across conversation
Token/cost limits	✅ Supported	Enforced per-request limits
Model optimization	✅ Supported	Can switch to cheaper models
Tool results compression	✅ Supported	TOON compression implemented
Dual LLM verification	✅ Supported	Works with optimization rules
Metrics and observability	✅ Supported	Full Prometheus metrics integration
Token estimation	✅ Supported	Automatic for streaming (usage is null)

Chat ✅

Feature	Status	Notes
Chat conversations	✅ Works	Full message history support
Model listing	✅ Works	Hardcoded model list (no API endpoint)
Model selection	✅ Works	Dropdown with all available models
Streaming responses	✅ Works	Real-time SSE streaming
Reasoning content	✅ Works	Displays thinking process separately
Error handling	✅ Works	MiniMax-specific error codes handled
API key management	✅ Works	Personal/team/org-wide key hierarchy
Conversation titles	✅ Works	Auto-generation using fast model
Token tracking	✅ Works	Estimated tokens logged to interactions

Testing Infrastructure

• ✅ E2E Test Coverage (5 test suites)
  • Tool invocation policies (tool-invocation.spec.ts)
  • Tool persistence (tool-persistence.spec.ts)
  • Tool result compression (tool-result-compression.spec.ts)
  • Model optimization (model-optimization.spec.ts)
  • Token cost limits (token-cost-limits.spec.ts)
• ✅ WireMock Stubs (13 mapping files in helm/e2e-tests/mappings/)
  • Model listing
  • Chat completions with various scenarios
  • Tool invocation test cases
  • Compression and optimization scenarios
  • Cost limit testing
• ✅ CI Configuration (.github/values-ci.yaml)

API Key Instructions

Obtaining an API Key

Visit MiniMax Platform and generate an API key (a minimum of $25 recharge is required to use the API)

Demo Video

minimax.mp4

Testing Done

• ✅ Chat conversation with MiniMax-M2.1-lightning
• ✅ Verified streaming response with curl to actual API
• ✅ Verified non-streaming response with curl to actual API
• ✅ Token estimation accuracy verified (input/output/reasoning tokens)
• ✅ Conversation title generation works
• ✅ Interactions logged with cost data
• ✅ Model listing shows all MiniMax models
• ✅ pnpm lint passes
• ✅ pnpm type-check passes

Documentation

Updated platform-supported-llm-providers.md

[London-Cat]

📊 Reputation Summary

User	Rep	Pull Requests	Activity	Assigned	Core Reactions
joeyorlando	⚡ 2089	85✅ 3🔄 12❌	100 issues, 50 comments	7	—
Rutetid	⚡ 48	2✅ 1🔄 1❌	0 issues, 17 comments	1	+1👍

---

How is the score calculated? Read about it in the Reputation Bot repository 🤖

[Rutetid]

Hey, I’ve got the MiniMax provider working, but I’m hitting a wall with the logs , it’s sometimes not showing up in the logs no matter what I try. Also, I noticed /profiles is gone, is that part of some shift? Would love some quick guidance on how to fix these issues!

[Copilot]

Pull request overview

Adds end-to-end MiniMax provider support across the proxy, chat UI, typed schemas/OpenAPI, pricing, and e2e tests.

Changes:

• Introduces MiniMax proxy routes + v2 adapter (streaming, tool calls, token estimation, error mapping).
• Adds MiniMax provider types/schemas and propagates the provider through shared constants, API client types, and OpenAPI.
• Adds chat UI integration, pricing migration, CI config, and e2e tests + WireMock mappings.

Reviewed changes

Copilot reviewed 60 out of 62 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
platform/shared/routes.ts	Adds MiniMax route IDs for proxy endpoints.
platform/shared/model-constants.ts	Registers minimax provider, display name, and model marker patterns.
platform/shared/hey-api/clients/api/types.gen.ts	Adds generated MiniMax request/response types and extends provider unions.
platform/shared/hey-api/clients/api/sdk.gen.ts	Adds generated SDK methods for MiniMax chat completions endpoints.
platform/shared/hey-api/clients/api/index.ts	Re-exports MiniMax SDK functions and types.
platform/shared/chat-error.ts	Adds MiniMax error type constants.
platform/helm/e2e-tests/mappings/minimax-tool-persistence.json	WireMock stub for tool persistence scenario.
platform/helm/e2e-tests/mappings/minimax-token-cost-limit-test.json	WireMock stub for token/cost limits scenario.
platform/helm/e2e-tests/mappings/minimax-models-list.json	WireMock stub for model listing (used by tests).
platform/helm/e2e-tests/mappings/minimax-model-optimization-with-tools.json	WireMock stub for optimization with tools.
platform/helm/e2e-tests/mappings/minimax-model-optimization-short.json	WireMock stub for optimization short prompt scenario.
platform/helm/e2e-tests/mappings/minimax-model-optimization-no-tools.json	WireMock stub for optimization without tools.
platform/helm/e2e-tests/mappings/minimax-model-optimization-long.json	WireMock stub for optimization long prompt scenario.
platform/helm/e2e-tests/mappings/minimax-model-optimization-disabled.json	WireMock stub for optimization disabled scenario.
platform/helm/e2e-tests/mappings/minimax-compression-enabled.json	WireMock stub for TOON compression enabled scenario.
platform/helm/e2e-tests/mappings/minimax-compression-disabled.json	WireMock stub for TOON compression disabled scenario.
platform/helm/e2e-tests/mappings/minimax-blocks-tool-untrusted-data.json	WireMock stub for tool invocation policy blocking scenario.
platform/helm/e2e-tests/mappings/minimax-allows-regular-after-archestra.json	WireMock stub for mixed tool call sequence scenario.
platform/helm/e2e-tests/mappings/minimax-allows-archestra-untrusted-context.json	WireMock stub for untrusted context ordering scenario.
platform/frontend/src/lib/llmProviders/minimax.ts	Adds UI interaction mapper for MiniMax interactions.
platform/frontend/src/lib/interaction.utils.ts	Wires MiniMax interaction type into dynamic interaction handling.
platform/frontend/src/components/proxy-connection-instructions.tsx	Adds MiniMax proxy connection instructions (base URL + label).
platform/frontend/src/components/chat/model-selector.tsx	Adds MiniMax logo provider mapping for model selector.
platform/frontend/src/components/chat-api-key-form.tsx	Adds MiniMax provider config (name/icon/link/placeholder).
platform/e2e-tests/tests/api/llm-proxy/tool-result-compression.spec.ts	Includes MiniMax in shared compression test suite.
platform/e2e-tests/tests/api/llm-proxy/tool-persistence.spec.ts	Includes MiniMax in shared tool persistence test suite.
platform/e2e-tests/tests/api/llm-proxy/tool-invocation.spec.ts	Includes MiniMax in shared tool invocation policy tests.
platform/e2e-tests/tests/api/llm-proxy/token-cost-limits.spec.ts	Includes MiniMax in shared token/cost limit tests.
platform/e2e-tests/tests/api/llm-proxy/model-optimization.spec.ts	Includes MiniMax in shared model optimization tests.
platform/backend/src/types/llm-providers/minimax/tools.ts	Adds Zod schemas for MiniMax tools + tool choice.
platform/backend/src/types/llm-providers/minimax/messages.ts	Adds Zod schemas for MiniMax messages/tool calls/reasoning details.
platform/backend/src/types/llm-providers/minimax/index.ts	Exposes MiniMax schemas/types including stream chunk typing.
platform/backend/src/types/llm-providers/minimax/api.ts	Adds Zod request/response/headers schemas for MiniMax.
platform/backend/src/types/llm-providers/index.ts	Exports MiniMax provider types from the provider index.
platform/backend/src/types/interaction.ts	Adds MiniMax to interaction request/response unions + discriminated union.
platform/backend/src/types/chat-api-key.ts	Allows minimax as a supported chat API key provider.
platform/backend/src/tokenizers/index.ts	Selects tokenizer for MiniMax (tiktoken).
platform/backend/src/tokenizers/base.ts	Extends ProviderMessage union to include MiniMax message shape.
platform/backend/src/server.ts	Registers MiniMax schemas in OpenAPI registry.
platform/backend/src/routes/proxy/utils/cost-optimization.ts	Adds MiniMax message typing for cost optimization utilities.
platform/backend/src/routes/proxy/utils/adapters/minimax.ts	Adds legacy metrics adapter helper for MiniMax usage extraction.
platform/backend/src/routes/proxy/routesv2/minimax.ts	Adds MiniMax unified proxy routes and optional passthrough proxying.
platform/backend/src/routes/proxy/adapterV2/minimax.ts	Implements MiniMax v2 adapter: request/response/stream, SSE parsing, token estimation, tool compression, error extraction.
platform/backend/src/routes/proxy/adapterV2/index.ts	Exports MiniMax adapter factory.
platform/backend/src/routes/index.ts	Registers MiniMax proxy routes in backend routing.
platform/backend/src/routes/features.ts	Adds minimaxEnabled flag to feature endpoint response.
platform/backend/src/routes/chat/routes.models.ts	Adds MiniMax model fetcher (hardcoded models) and provider api key fallback.
platform/backend/src/routes/chat/routes.chat.ts	Adds MiniMax to smart default model selection.
platform/backend/src/routes/chat/errors.ts	Adds MiniMax error parsing + error-code mapping.
platform/backend/src/models/optimization-rule.ts	Adds MiniMax key to optimization rule provider maps.
platform/backend/src/llm-metrics.ts	Adds token reporting for MiniMax responses when usage exists.
platform/backend/src/database/migrations/meta/_journal.json	Registers the new MiniMax token price migration.
platform/backend/src/database/migrations/0131_add_minimax_token_prices.sql	Adds MiniMax token pricing rows.
platform/backend/src/config.ts	Adds MiniMax proxy and chat config knobs (base URLs, API key).
platform/backend/src/clients/models-dev-client.ts	Maps models.dev minimax to internal provider and source matching.
platform/backend/src/clients/llm-client.ts	Adds MiniMax provider detection, fast model, and model creators (direct + proxied).
platform/backend/src/clients/dual-llm-client.ts	Adds a MiniMax dual LLM client using OpenAI SDK compatibility.
docs/pages/platform-supported-llm-providers.md	Documents MiniMax provider setup and behavior.
docs/openapi.json	Updates generated OpenAPI with MiniMax schemas and endpoints.
.vscode/settings.json	Adds VS Code setting for Postman dotenv detection notification.
.github/values-ci.yaml	Adds MiniMax env vars for CI/e2e configuration.

[Rutetid]

Applied all suggestions by copilot!

[joeyorlando]

hi there @Rutetid 👋 I just merged #2610 which simplified a few parts regarding adding a new LLM provider - this introduced a few merge conflicts in your PR, do you mind rebasing off of latest main + addressing merge conflicts? I will review once that is done!

Additionally, can you run pnpm check:ci locally and address any issues + push up? 🙏

[Rutetid]

Hi @joeyorlando I've rebased onto the latest main to include the #2610 refactors. I've also confirmed that pnpm check:ci is passing with all tests green. Ready for final review!

[joeyorlando]

we don't need a database migration

[Rutetid]

alright ill remove it

[Rutetid]

I removed the migration , and now the values will default to $30 for cheaper models and $50 for others as specified in the file backend/src/default-model-prices.ts