v2.5 Updates

Overview

Version 2.5 introduces a large number of new features, including an extended Functions engine, an OpenAI-compatible API, on-device MCP support, and more. Built on the Anyone parallel-inference architecture, streaming is now the default—and only—response mode.

Massive Functions expansion

New OpenAI-compatible `/chat/completions` endpoint

On-device MCP (Model Context Protocol) support

Streaming is now the default & only response mode

Role-based encapsulation for simpler integration

Authentication

Every request must include the following headers:

api_key: 32-character string

org_id: 6-character string

Functions Reference

v2.5 dramatically expands the Functions engine, giving the AI the ability to call external APIs and perform real-world tasks.

Available Functions

#	Function	Purpose	Trigger Example
1	`get_weather`	Real-time weather for the next 5 days	“What’s the weather like?”
2	`get_news`	Latest headlines & hot events	“Any news today?”
3	`get_vision`	Image recognition / processing	“What’s in this picture?”
4	`dance`	Trigger a character dance (ShuBan only)	“Dance for me!”
5	`search_internet`	Tavily AI powered web search	“Search for the latest GPU benchmarks”
6	`search_baidu`	Baidu search optimised for Chinese	“百度一下今天的股市”
7	`read_url`	Extract main text from any URL	“Summarise this article”
8	`book_house_service`	Book cleaning, cooking, errands (Aiguo only)	“I need a cleaner tomorrow”
9	`car_sales`	EV model & price lookup (SiYou only)	“Show me Tesla Model Y prices”
10	`tarot`	Tarot card draw & reading	“Draw three cards for my career”
11	`bazi`	Ba-Zi astrology via birth data	“Analyse my birth chart”
12	`search_memory`	Search historical chat context	“What did I ask last Friday?”
13	`lingtong_action`	Ling-tong special action (Ling-tong only)	“Perform action #3”

Configuration

Intro

{
  "functions": ["search_internet", "tarot", "bazi"],
  "voice_id": "wukong"
}

Usage

Each role can enable multiple functions.

The system auto-selects the best function(s) per turn.

nctions can be chained.

me require extra parameters (e.g., birth date for bazi).

OpenAI-Compatible API

Role-based Design Philosophy

Unlike the traditional OpenAI library, we use a role-based wrapper:

No messages array needed – just send a prompt string.

Auto context management – the system keeps and uses conversation history automatically.

Role-driven interaction – pick the character you want to talk to with the model parameter.

`/chat/completions` Endpoint

POST /chat/completions

Authentication

Request Body

{
    "model": "NPC ID",
    "prompt": "User Input",
    "stream": true,
    "lang": "default",
    "emoji_mode": true,
    "audio_mode": true,
    "emotion_status": false,
    "tools": ["search_internet", "tarot"],
    "tool_call": false,
    "function_id": null
}

Parameter Reference

Args	Type	Must	Default Value	Intro
model	string	Yes	-	NPC ID
prompt	string	Yes	-	User Input
lang	string	No	"default"	Language
emoji_mode	boolean	No	true	Enable emoji
audio_mode	boolean	No	true	Enable auto break
emotion_status	boolean	No	false	Enable emotion status
tools	array	No	[]	Avaliable tools list
tool_call	boolean	No	false	Is tool calling
function_id	string	No	null	Function ID

Response Format

Adopt the standard OpenAI streaming response format.

{
    "id": "chatcmpl-xxx",
    "object": "chat.completion.chunk",
    "created": 1234567890,
    "model": "NPC ID",
    "choices": [
        {
            "index": 0,
            "delta": {"content": "response content"},
            "finish_reason": null
        }
    ]
}

Sample

On-Device MCP Support

What is On-Device MCP?

On-device MCP (Model Context Protocol) allows you to embed MCP directives directly within the messages you send, working in conjunction with server-side functions.
Refer to the TCP interface MCP documentation for details.
Alternatively, you can use the plain HTTP interface and pass MCP capabilities through functions.

Streaming Response Features

Anyone Parallel-Inference Architecture

Streaming by default: All responses are streamed and cannot be disabled

Parallel inference: Multiple inference tasks run concurrently

Real-time: Results are emitted while still being computed, providing the best user experience

Response Formats

Legacy interface: Server-Sent Events (SSE)

OpenAI-compatible interface: Standard OpenAI streaming format

Error Codes

Error Code	Intro
200	success
400	args error
401	auth error
403	forbbiden
404	not found
429	too many
500	internal error
601	login ex

Changelog

v2.5 Major Updates:

Significantly More Functions: Expanded from 4 to 13+

OpenAI-Compatible API: New /chat/completions endpoint

On-Device MCP Support: MCP directives can now be embedded directly in messages

Streaming Response Optimized: Default streaming powered by parallel-inference architecture

Role-Based Encapsulation: Simplified calls with automatic context management

Legacy Cleanup: Removed fast_mode, route, and related parameters

Deprecated Features:

fast_mode parameter

route parameter

/prepare_fast_mode endpoint

Non-streaming response mode

Important Notes

Credential Security: Keep your API KEY and ORG ID safe

Request Limits: Prompt length capped at 1,500 characters

Rate Limiting: All endpoints enforce request-rate limits

Function Permissions: Some functions require extra authorization

Streaming Only: Only streaming responses are supported; they cannot be disabled

Compatibility: New integrations should use the OpenAI-compatible endpoints

Overview#

Massive Functions expansion#

New OpenAI-compatible /chat/completions endpoint#

On-device MCP (Model Context Protocol) support#

Streaming is now the default & only response mode#

Role-based encapsulation for simpler integration#

Authentication#

Functions Reference#

Available Functions#

Configuration#

OpenAI-Compatible API#

Role-based Design Philosophy#

/chat/completions Endpoint#

Authentication#

Request Body#

Parameter Reference#

Response Format#

Sample#

On-Device MCP Support#

What is On-Device MCP?#

Streaming Response Features#

Anyone Parallel-Inference Architecture#

Response Formats#

Error Codes#

Changelog#

v2.5 Major Updates:#

Deprecated Features:#

Important Notes#

Overview

Massive Functions expansion

New OpenAI-compatible `/chat/completions` endpoint

On-device MCP (Model Context Protocol) support

Streaming is now the default & only response mode

Role-based encapsulation for simpler integration

Authentication

Functions Reference

Available Functions

Configuration

OpenAI-Compatible API

Role-based Design Philosophy

`/chat/completions` Endpoint

Authentication

Request Body

Parameter Reference

Response Format

Sample

On-Device MCP Support

What is On-Device MCP?

Streaming Response Features

Anyone Parallel-Inference Architecture

Response Formats

Error Codes

Changelog

v2.5 Major Updates:

Deprecated Features:

Important Notes