Interface and Capability Overview

CoreRouter doesn't have just one type of interface. Before integrating, first determine which protocol your client will call, then select the corresponding Base URL, Endpoint, and model capabilities.

Quick Decision

What You Want to Do	Priority Use	Base URL	Is It Universal
Regular chat, multi-turn conversation, function calling, vision understanding	/v1/chat/completions	https://api.corerouter.cloud/v1	Common universal entry
Legacy text completion	/v1/completions	https://api.corerouter.cloud/v1	Only for legacy applications
Codex, Agent, Responses workflow	/v1/responses	https://api.corerouter.cloud/v1	Conditional support, requires model with Responses capability
Agent standalone search	/v1/alpha/search	https://api.corerouter.cloud/v1	Conditional support, requires search-compatible channel and model
Claude Code, Anthropic-style applications	/v1/messages	https://api.corerouter.cloud	Conditional support, requires client using Anthropic Messages format
Gemini-style applications	/v1beta/models/{model}:generateContent	https://api.corerouter.cloud	Conditional support, requires client using Gemini request format
Embeddings / RAG	/v1/embeddings	https://api.corerouter.cloud/v1	Requires Embeddings model
Rerank / Reranking	/v1/rerank	https://api.corerouter.cloud/v1	Requires Rerank model or channel
Moderations / Content moderation	/v1/moderations	https://api.corerouter.cloud/v1	Requires moderation model or available default moderation model
Image generation and editing	/v1/images/generations, /v1/images/edits	https://api.corerouter.cloud/v1	Requires image model
Speech recognition, translation, TTS	/v1/audio/...	https://api.corerouter.cloud/v1	Requires audio or TTS model
Realtime WebSocket	/v1/realtime	wss://api.corerouter.cloud/v1/realtime	Conditional support, requires Realtime model
Video tasks	/v1/videos, /v1/video/generations	https://api.corerouter.cloud/v1	Conditional support, requires video task model and task channel
Midjourney-style image tasks	/mj/submit/...	https://api.corerouter.cloud	Conditional support, requires corresponding image task channel
Generic asynchronous tasks	/v1/tasks/{key}	https://api.corerouter.cloud/v1	Advanced interface, typically used by plugins or integrated clients

Endpoint List

Endpoint	Method	Request Format	Description
/v1/models	GET	No body	Query models visible to current API Key. Results may be affected by user groups, Key restrictions, and billing configuration.
/v1/models/{model}	GET	No body	Query single model.
/v1beta/models	GET	No body	Gemini-style model list, returns Gemini-compatible models visible to current API Key.
/v1beta/openai/models	GET	No body	OpenAI-style model list compatible path for Gemini scenarios.
/v1/chat/completions	POST	OpenAI Chat Completions JSON	Most commonly used chat entry.
/v1/completions	POST	OpenAI legacy Completions JSON	Legacy application compatibility entry, new applications prioritize Chat Completions.
/v1/responses	POST	OpenAI Responses JSON	Commonly used by Agent / Codex, only Model IDs supporting Responses can be used.
/v1/responses/{response_id}	GET	No body	Query background Responses result.
/v1/responses/compact	POST	Responses compact request	Advanced capability, typically automatically called by Agent tools; requires channel support for compact capability.
/v1/alpha/search	POST	Client native search JSON	Standalone search compatibility entry, only channels and models configured with corresponding search capability available.
/v1/messages	POST	Anthropic Messages JSON	Used by Claude Code and Anthropic-style clients.
/v1beta/models/{model}:generateContent	POST	Gemini JSON	Gemini-style generation interface.
/v1beta/models/{model}:streamGenerateContent	POST	Gemini JSON	Gemini-style streaming interface.
/v1beta/models/{model}:embedContent	POST	Gemini Embedding JSON	Gemini-style single vector interface.
/v1beta/models/{model}:batchEmbedContents	POST	Gemini Embedding JSON	Gemini-style batch vector interface.
/v1/embeddings	POST	OpenAI Embeddings JSON	Used for RAG and vector retrieval.
/v1/rerank	POST	Rerank JSON	Rerank candidate documents by relevance.
/v1/moderations	POST	OpenAI Moderations JSON	Content moderation.
/v1/images/generations	POST	JSON	Image generation.
/v1/images/edits	POST	multipart/form-data or JSON	Image editing.
/v1/edits	POST	JSON	Legacy image editing compatibility path, prefer /v1/images/edits.
/v1/audio/transcriptions	POST	multipart/form-data	Audio to text.
/v1/audio/translations	POST	multipart/form-data	Audio translation to English or model's default target language.
/v1/audio/speech	POST	JSON	Text to speech.
/v1/realtime?model={model}	GET WebSocket	WebSocket event JSON	Real-time voice/text interaction, not regular HTTP interface.
/v1/videos	POST	JSON or multipart/form-data	OpenAI-style video task submission.
/v1/videos/{task_id}	GET	No body	OpenAI-style video task query.
/v1/videos/{task_id}/content	GET / HEAD	No body	Get video task output, availability depends on task channel.
/v1/video/generations	POST	JSON	Compatible video task submission path.
/v1/video/generations/{task_id}	GET	No body	Compatible video task query path.
/v1/videos/{video_id}/remix	POST	JSON	Regenerate based on existing video, requires channel support.
/v1/tasks/{key}	POST	JSON or multipart/form-data	Generic task submission, key is task plugin or task type identifier.
/v1/tasks/{task_id}	GET	No body	Generic task query.
/v1/tasks/{task_id}/artifacts	GET	No body	Query task artifact list.
/v1/tasks/{task_id}/artifacts/{artifact_key}/content	GET / HEAD	No body	Download or probe task artifact content.
/mj/submit/imagine	POST	Midjourney-style JSON	Image task submission, prompt required.
/mj/task/{id}/fetch	GET	No body	Query image task status and result.

Don't Mix Request Formats

Different protocols have different JSON structures; can't just change URL without changing request body.

Protocol	User Message Field
Chat Completions	messages: [{ "role": "user", "content": "..." }]
Responses	input: "..." or structured input array
Anthropic Messages	messages, max_tokens, tool fields use Anthropic format
Gemini	contents: [{ "parts": [{ "text": "..." }] }]
Embeddings	input: "..." or string array
Rerank	query and documents

How to Diagnose Common Errors

Phenomenon	Common Cause	Solution
Chat works, Responses doesn't	Current Model ID only supports Chat Completions	Switch to model marked as supporting Responses in console
Client path shows /v1/v1	Base URL and client auto-append path conflict	Change Base URL to root address or address with /v1, choose one
/v1/models has results, calling model errors	Key can see model list, but target model permissions, channel, or capability don't match	Switch model or check Key restrictions and account groups
messages is required	Chat Completions request body missing messages	Use Chat format, don't pass Responses' input
input is required	Responses, Embeddings, or Moderations missing input	Add input per corresponding interface
query is empty or documents is empty	Rerank request missing query or candidate documents	Add query and non-empty documents
Realtime directly using curl POST	Realtime is WebSocket, not HTTP JSON POST	Use WebSocket client to connect wss://.../v1/realtime?model=...
Standalone search returns unsupported	Current channel or model has no search compatibility capability	Switch to model supporting search, or use client-supported search tools in Responses workflow

Recommended Integration Order

1. Use /v1/models to confirm API Key is available.
2. Use /v1/chat/completions to verify basic chat.
3. Verify target interface per actual scenario, e.g., Responses, Embeddings, Rerank, audio, or video.
4. Then fill the same set of Base URL, API Key, and Model ID into SDK, client, or plugin.

Interfaces Not Recommended to Promise Support

The following OpenAI legacy or management interfaces are not universal relay call entries. Even if they appear in client functionality, don't promise availability by default:

Interface Type	Description
/v1/files	File upload, file content reading, and other file management interfaces are not general model call entries.
/v1/fine-tunes	Legacy fine-tuning interface is not a regular integration capability.
/v1/assistants, /v1/threads, /v1/runs	Assistants legacy resources and thread interfaces not opened as universal relay entries.
/v1/batches, /v1/vector_stores	Batch processing and vector store management interfaces not opened as universal relay entries.
POST /v1/messages/count_tokens	Anthropic token counting helper interface currently not a public integration capability.
/v1/images/variations	Image variation interface not equivalent to image generation or editing, requires separate confirmation before integration.
DELETE /v1/models/{model}	Model deletion belongs to official management interface semantics, not suitable as relay station user capability.

If a client strongly depends on these interfaces, first confirm whether it allows disabling related functionality, or switch to clients that only depend on Chat Completions, Responses, Anthropic Messages, Gemini, Embeddings, Rerank, images, audio, video tasks — these documented interfaces.