Awesome AI AgentsVirtual Assistants

Bitterbot-AI/bitterbot-desktop

⭐ 2462 TypeScript repository created 2026-03-28

Bitterbot is a local-first personal AI that runs on personal devices instead of being a stateless wrapper around an LLM API. It ships as a Node 22 project managed with pnpm: a setup script installs Chromium, ffmpeg and ripgrep, an onboarding wizard walks through model authentication, memory embeddings, web search, channels, wallet and workspace setup, and it then starts the gateway and Control UI automatically. The gateway is a WebSocket backend on port 19001, the Control UI is a browser dashboard on port 5173 where the operator chats, views dreams, manages skills and monitors the agent, and a peer-to-peer orchestrator sidecar is spawned by the gateway. A CLI can send single messages to a named agent, and watch modes are provided for development on the source. The memory design is described as a cognitive architecture rather than a vector store with a retrieval step. Knowledge Crystals decay along Ebbinghaus forgetting curves, and a consolidation pipeline runs every thirty minutes to apply hormonal decay, merge chunks, forget low-importance material and enforce governance. Three simulated neuromodulators shape behaviour in real time: dopamine for achievement, cortisol for urgency and oxytocin for bonding, and eight response dimensions such as warmth, energy, focus, playfulness, verbosity, curiosity, assertiveness and empathy are computed from that blend each turn. A curiosity engine maps gaps, contradictions and semantic frontiers through a five-component reward function whose alpha parameter shifts from density-seeking to frontier-seeking as the agent matures. Proactive recall injects identity and directive memories every turn at no LLM cost, and a canonical facts ledger keeps a small always-injected layer of ground truth that bypasses similarity search, with restatement strengthening a fact and contradiction superseding it while keeping its history. A typed knowledge graph records the people, projects and things in the life of the user. The agent also browses the web, runs code, talks over WhatsApp, and packages learned skills for trade with other agents for USDC.

https://github.com/Bitterbot-AI/bitterbot-desktop

personal-assistantagent-memorylocal-firstdesktop-agentp2p

Also in Virtual Assistants

mem0ai/mem0

Mem0 is an open-source memory layer that enhances AI agents and assistants with personalized, adaptive memory capabilities for improved context-aware interactions across various applications.

CherryHQ/cherry-studio

Cherry Studio is a cross-platform desktop client that supports multiple large language model providers, offering extensive AI assistant features, document processing, and practical productivity tools in an open-source environment.

tinyhumansai/openhuman

Local-first personal AI harness combining a persistent memory brain, multi-agent orchestration, and autonomous web and data research.

khoj-ai/khoj

Khoj is a self-hostable personal AI platform that transforms any local or online large language model into a customizable, autonomous AI assistant for research, automation, and document retrieval.

kortix-ai/suna

Suna is an open-source generalist AI assistant that automates real-world tasks through natural conversation, combining browser automation, file management, web crawling, and API integrations in a secure, modular architecture.

gptme/gptme

gptme is a versatile personal AI assistant that operates in your terminal, equipped with tools for coding, file editing, web browsing, and vision, providing a powerful local alternative to popular AI agents for diverse knowledge work.

CherryHQ/cherry-studio-app

Cherry Studio App is the official mobile application that enables powerful AI Large Language Models interaction on iOS and Android devices, supporting multiple LLM providers and offering tools for conversation management and data migration.

cosmicstack-labs/mercury-agent

Personal AI agent in TypeScript that runs continuously from CLI, Telegram or web with permission-gated tools, SQLite-backed long-term memory, token budgets and installable skills.