User repository on GitHub
🥰 Building AI-based conversational avatars lightning fast ⚡️💬
- Forks
- 68
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Integrations & Adapters
Aiavatarkit
A Python speech-to-speech avatar framework: swappable VAD, STT, LLM, and TTS modules behind one streaming pipeline, with WebSocket, telephony, LINE, and OpenAI-compatible channel adapters on top.
- Stars
- 691
- Language
- Python
- License
- APACHE-2.0
- Published by
- Community
- Author
- uezo
- Last verified
- Oct 2026
About this project
🥰 Building AI-based conversational avatars lightning fast ⚡️💬. Talking avatars in your app — a character in your web or mobile app that speaks, reacts with expressions and animations. Interactive signage and virtual store staff — reception, wayfinding, and product guidance that answers out loud while putting images and UI on the screen. Companion devices — Raspberry Pi, M5Stack, StackChan, and other hardware; add voice conversation to anything.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- A single pipeline streams voice activity detection into transcription, the LLM, and speech synthesis, each stage swappable.
- Adapters own transport only, and several can attach to a single pipeline instance so a conversation survives channel changes.
- Turn-taking gates include a Jev gate alongside session allowance, multiple gates, and a custom option; turn-end gates add semantic VAD.
- Ships an admin panel for config, logs, metrics, and evaluation, plus tracing and mid-sentence guardrail interruption.
Watch out
The sub-second latency figure is the project's own measurement with no hardware, sample size, or method described, so treat it as a claim rather than a budget. The default application needs an OpenAI key and a reachable VOICEVOX-compatible server before you configure anything, plus Python 3.11 or newer, and first launch asks interactively whether to install optional semantic-VAD extras. The README warns that blog walkthroughs may predate the current release and suggests pinning an older version to match them. No license is stated in the visible portion, and nothing covers cost: most stages are metered hosted APIs, while local alternatives appear only as configuration options. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
The README does not say. It lists hosted services for every stage and local or self-hosted options such as VOICEVOX, AivisSpeech, Style-Bert-VITS2, SpeechGateway, and LM Studio, but states no hardware requirements, concurrency limits, or which combinations run without a GPU. Its only quantitative claim is under one second from end of speech to first audio, measured by the project.
Stages can be swapped, not skipped. Voice activity detection, turn-end gates, turn-taking gates, speech-to-text, the LLM, and speech synthesis are each a module with several implementations, and the turn-taking gate list includes a Jev option. The default application assumes the whole speech-to-speech path, and the customization recipe still passes an OpenAI key for transcription even when you swap the LLM to Claude.
That is the stated design intent. Adapters wrap the pipeline for one channel and own only transport concerns, and multiple adapters can attach to the same pipeline instance, so the README describes hanging up a call and continuing in a messaging app. Available channel adapters are WebSocket, HTTP with server-sent events, a LINE bot, Twilio voice and SMS, Asterisk, an OpenAI-compatible endpoint, and a recognition-only server.
The README says tools load only when needed so a large catalog does not confuse the model, and slow tools do not stall the conversation: they run in the background, or the avatar replies from a template instead. Tool calls and MCP are supported at the LLM stage. How a backgrounded result is folded back into the conversation once it completes is not described.
Python 3.11 or newer, an OpenAI API key, and a reachable VOICEVOX-compatible server at its default URL. Install with pip install aiavatar, export OPENAI_API_KEY, then run the aiavatar command; the avatar opens at 127.0.0.1:8000 and the Admin Panel at /admin/. First launch asks whether to install the optional Namo Turn semantic-VAD extras.
It is one of the turn-end gates and one of the turn-taking gates, alongside Smart Turn, Namo Turn, OpenAI Decisions, LLM-based and custom gates. Turn boundaries and the wakeword are judged by meaning rather than silence or keywords, so a thinking pause does not cut the user off and a backchannel does not interrupt the avatar.
The Admin Panel covers config, logs, metrics and evaluation, and the README says you can retune a running pipeline without restarting. Guardrails run in parallel and can interrupt the avatar mid-sentence to correct what it just said, and Langfuse tracing is built in.
More Jev resources like this
1.1kGitHub stars
A section-by-section guide to agent harnesses - loop, tool runtime, permissions, hooks, subagents, skills, memory, background work - using Claude Code, mini-swe-agent, Hermes Agent, and deepseek-harness as worked examples.
Jev Integrations & Adapters#community#jevbest
61GitHub stars
The model-integration layer for the Milvus vector database, shipping adapters for embedding and reranker models from hosted providers, Jev among them, plus local backends.
Jev Integrations & Adapters#community#jevbest
Community
milvus-io/milvus-modelPython
milvus-io/milvus-modelPython430GitHub stars
Serves Jev-style decision checkpoints through vLLM on Linux, with an MLX preview for Apple Silicon, exposing Choice, Noul and Score answers over HTTP.
Jev Integrations & Adapters#community#jevbest
Community
mode-io/vllm-jevPython
mode-io/vllm-jevPython41GitHub stars
Typed Python planner runtime that turns batched Jev action distributions into executable probability trees and graphs for finite decision tasks.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
Chuf-H/jev-treePython
Chuf-H/jev-treePython25GitHub stars
Agent runtime with a terminal UI and a messaging gateway that keeps memory, self-built skills, session search, scheduling, and subagents across Telegram, Discord, Slack, WhatsApp, Signal, and CLI.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
wundercorp/lokiPython
wundercorp/lokiPython51GitHub stars
GraphRAG layer that runs a 4-phase ingest-to-answer pipeline over Neo4j, Memgraph, AGE or Kuzu using Laya or Jev decisions instead of per-hop LLM calls.
Jev Integrations & Adapters#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.