AI Dev Impact Lab JA
← Rankings · AI-RELATED REPO

hsergiu/convers

Local voice agent (eng) with barge-in. Three layer memory, browser as mic over LAN, ~1.9 s to first audio on a 4070 Super

Python APACHE-2.0 GitHub ↗
★ 0
stars
70
AI relevance
50
solo dev
0
tool sigs

SUMMARY AI summary by gpt-5-mini

convers is a local-first conversational agent for developers who want a privacy-preserving, interruptible voice assistant running entirely on localhost. It provides real-time, full‑duplex turn-taking with true barge‑in (TTS cancels mid‑word), and a browser HTTPS+WebSocket front end so a phone can act as mic and speaker. Key features: - Pluggable providers (VAD/STT/LLM/TTS) via Python entry points. - Three-layer memory: curated profile, SQLite RAG with precomputed embeddings for session-spanning recall, and a short in-process rolling buffer for immediate context. - Streaming pipeline: mic → VAD → STT → LLM → TTS with a cancel chain for interruptions. - Uses llama-server for chat (with --n-cpu-moe) and Ollama for embeddings; Rust audio extension for low-latency playback. - Clear runtime invariants and modular architecture for maintainability.

DETECTED Detected AI stack

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.

🧠 LLM providers (1)
Ollama
🤖 LLM models (1)
Llama
💾 Databases (1)
SQLite

GitHub Topics

#llm #local-agent #python #speech-to-text #text-to-speech #voice-agent #voice-assistant #whisper

Language breakdown (by bytes)

HTML
2.4%
Rust
2%
Shell
0.8%
Python
94.7%

Owner

Account
hsergiu
Type
User
Followers
5

Dates

Created on GitHub 2026-05-09
Last push 2026-05-09
First seen here 2026-05-09
Last fetched 2026-05-09 18:12

Similar repos (same language)

bgzhang1/sw2api

Reverse proxy for your ai quota from the SW platform.

Python 19 AI 45
dea6cat/2b-agent

This llm Agent was created based on necessity for one that could simply use local models without making them hallucinate and keep them focus

Python 1 AI 75
Zuboy/Carbon-Cost-Optimizer-Agentic-AI

An AI agent that decides where and when to run ML training jobs to minimize cost and carbon emissions, then launches them autonomously — exposed entirely through MCP tools.

Python 1 AI 75
FuchaZ/lm-studio-vision-bridge

给纯文本 AI agent 装上眼睛——通过 LM Studio 本地视觉模型提供 MCP 图片识别服务

Python 1 AI 45
Shrutirowlo/AI-Video-Agent

AI-powered video generation agent that automates script analysis, scene planning, and video creation.

Python 1 AI 75
Shivang0/social-vision

Paste an Instagram/TikTok/YouTube/X link and Claude watches it for you — transcribes audio, reads on-screen text & visuals, explains what it's about. Local, cross-platform Claude Code plugin.

Python 1 AI 70
markusbegerow/claude-fuer-oeffentliche-verwaltung

⚠️ Experimentelle Skill-Sammlung für Digitalisierung, Fachverfahren und Organisation der deutschen öffentlichen Verwaltung (OZG, FIM, FIT-Connect, KI-Einsatz, eIDAS, BSI-IT-Grundschutz, DSGVO) – bitte testen, Issues/PRs willkommen! Keine Rechts-/Datenschutzberatung, keine verbindliche Behördenentscheidung. Keine Bürger-/Mitarbeiterdaten im Repo.

Python 1 AI 75
knowledgegut/course-distiller

把有權使用的工作坊/課程回放消化成套品牌色的步驟式學習 PDF;最後的重組交還人腦。A Claude Code skill by 知識包小腸.

Python 1 AI 45