Skip to content

plugins.md

Latest commit

 

History

History
83 lines (57 loc) · 3.81 KB

plugins.md

File metadata and controls

83 lines (57 loc) · 3.81 KB
Raw

Available plugins

This package ships six plugins. Five are backed by a shared OpenAI Chat Completions client (ovos_openai_plugin.api.OpenAIChatCompletions); the RAG memory plugin instead talks to an OpenAI-compatible vector-stores search API.

ovos-chat-openai-plugin: Chat engine

  • Entry point group: opm.agents.chat
  • Class: ovos_openai_plugin.chat.OpenAIChatEngine (ChatEngine)

A multi-turn conversational agent for ovos-persona. Implements:

  • continue_chat(messages, ...) -> AgentMessage: full response
  • stream_tokens(messages, ...): raw token stream (not TTS-safe)
  • stream_sentences(messages, ...): complete sentences, buffered via sentence-stream (TTS-safe)
  • get_response(utterance, ...): convenience one-shot helper (from the base class)

System prompt handling is controlled by system_prompt and allow_system_prompts (see configuration).

Tool / function calling

OpenAIChatEngine sets supports_tools = True. Pass ToolBox object(s) (or OpenAI tool dicts) to continue_chat(..., tools=...). The engine exposes them to the model. When the model requests a tool, it returns an assistant AgentMessage whose tool_calls (a list of ToolCall) are populated. Feed the results back as MessageRole.TOOL messages and call again to continue. The ovos-native-toolcall-loop engine drives this loop for you. This plugin only needs to be the brain.

ovos-openai-rag-memory-plugin: RAG memory

  • Entry point group: opm.agents.memory
  • Class: ovos_openai_plugin.rag_memory.PersonaServerRAGMemory (AgentContextManager)

A persona memory plugin: before each turn, searches a vector store on an OpenAI-compatible server and injects the retrieved chunks into the conversation context that the chat engine sees. See RAG memory for the full configuration reference, how to build the vector store it searches, and its failure modes.

ovos-summarizer-openai-plugin: Summarizer

  • Entry point group: opm.agents.summarizer
  • Class: ovos_openai_plugin.summarizer.OpenAISummarizer (SummarizerEngine)

Summarizes a document via summarize(document, lang=None) -> str. The prompt is built from prompt_template (must contain {content}).

ovos-translate-openai-plugin: Translator

  • Entry point group: opm.lang.translate
  • Class: ovos_openai_plugin.translate.OpenAITextTranslator (LanguageTranslator)

translate(text, target=None, source=None) -> str. Language codes are expanded to human-readable names (via langcodes) before being inserted into the prompt. If target is omitted, the configured lang from OVOS configuration is used. If source is omitted, the model auto-detects it.

ovos-lang-detect-openai-plugin: Language detector

  • Entry point group: opm.lang.detect
  • Class: ovos_openai_plugin.translate.OpenAITextLangDetector (LanguageDetector)

detect(text) -> str returns a standardized BCP-47 tag. detect_probs(text) -> Dict[str, float] returns {lang: 1.0}.

ovos-dialog-transformer-openai-plugin: Dialog transformer

  • Entry point group: opm.transformer.dialog
  • Class: ovos_openai_plugin.dialog_transformers.OpenAIDialogTransformer (DialogTransformer)

Rewrites dialog text just before TTS. transform(dialog, context) -> (text, context). The rewrite prompt comes from context["prompt"] or the configured rewrite_prompt. If neither is set, the dialog is returned unchanged.

Demo persona

The opm.plugin.persona entry point Remote Llama exposes a ready-made persona (ovos_openai_plugin.LLAMA_DEMO) pointed at a public OpenAI-compatible LLama server.

← Configuration reference · Home · Persona integration →