Project Vibe
Vibe is a local-first AI orchestration system built to research, code, automate, and act.
It combines specialized agents, real computer control, hybrid memory, and sandboxed execution to help power users complete complex tasks from start to finish. Think of it as a Jarvis-style assistant experience grounded in specialized agents, real execution, and user-controlled safety boundaries.
Not a Chatbot. Not a Wrapper.
Most AI assistants stop at generating answers. Vibe is built as a multi-agent orchestration system that can research, code, use software, remember context, and carry work through with specialized agents coordinating behind the scenes.
LLM-Agnostic
Gemini, OpenAI, Anthropic, xAI, OpenRouter, Ollama — per agent
Real Desktop Control
Vision + mouse/keyboard — not just generating instructions
Runs Locally
Your data stays on your machine — secrets never leave .env
Fully Extensible
Plugin Development Kit for custom agents and capabilities
6 Agents. One Orchestrator.
Vibe uses specialized agents that work together so tasks can be routed to the model, toolset, and execution path that fit them best.
The Brain
Managerial Agent
Plans, routes, and synthesizes work across the system. Breaks goals into dependency-aware subtasks, delegates execution, recovers from failures, and keeps progress moving without losing the thread.
Information Specialist
Research Agent
Handles grounded search, deep research, URL analysis, local file reading, and research synthesis. It can also turn findings into deliverables in multiple formats, including PDF-style papers and structured writeups.
Full-Stack Developer
Coder Agent
Writes, debugs, and refactors code with direct filesystem access. For heavier tasks, it can execute scripts instead of wasting tokens on step-by-step round trips.
Vibe's Hands
Computer Use Agent
Handles real desktop interaction through vision, mouse, and keyboard control. Users can choose direct execution or isolated sandboxed paths depending on the task and risk level.
Visual Creator
Image Agent
Generates images and visual assets so creative work can stay inside the same workflow instead of breaking out into separate tools.
Quality Gate
Code Review Agent
Reviews code for quality, efficiency, and safety so generated work is checked before it is treated as done.
Built Different
The architecture matters because it changes what Vibe can actually help you do in practice.
Use Any Model
Route different agents to different models so research, coding, reasoning, and execution can each use the best fit.
6-Lane Hybrid Memory
Grep for code-level context, temporal for recent events, semantic for cross-session recall, multimodal for documents and assets, a curated fact store, and persistent files — all queried in parallel and fused together.
Programmatic Tool Calling
For data-heavy or repetitive work, Vibe can write and execute scripts instead of burning tokens on step-by-step generation.
Agent Collaboration
Specialized agents work together so tasks can be routed to the toolset, model, and execution path that fit them best.
Sandboxed Execution
Users can choose how computer use runs, including direct control and isolated sandboxed execution paths.
Plugin Development Kit
Vibe is extensible by design, with a plugin system for adding custom capabilities, workflows, and integrations.
Safety-First Architecture
Vibe is built with layered safety controls because an AI that can act needs stronger guardrails than a chatbot. It combines content-level blocking, intent-aware verification, configurable access boundaries, confirmation gates, and isolated execution paths for riskier work.
Path Allowlist and Blacklist — directory-level controls for protecting sensitive areas of the system
Two-Layer Safety Validation — content scanning plus intent-aware verification before risky execution proceeds
Safe Mode and Sandboxed Execution — from read-only operation to isolated execution paths based on user control and task risk
User Confirmation Gates — sensitive actions can pause for explicit authorization instead of assuming intent
ENV-Only Keys and Log Redaction — secrets stay referenced through configuration while sensitive values are scrubbed before logs hit disk
A Native Desktop App
Vibe ships as an Electron + Python application — a React GUI connected to a Python backend via ZMQ message bus. Real-time event feed, session dashboard, per-agent configuration, inline code editor, and live CUA status — all in one window.
6
Specialized Agents
6+
LLM Providers
3
Memory Tiers
6
Safety Layers
Interested in Vibe?
Vibe is in alpha and under active development. Get in touch to learn more, explore the direction, or talk multi-agent systems with us.