> ## Documentation Index
> Fetch the complete documentation index at: https://docs.unify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Overview
> An AI agent you can steer while it works
## What is Unity?
Unity is an AI agent you can interrupt mid-task, redirect without restarting, talk to while it works, and run several things on at once. It's voice-native, with memory that compounds over time.
## What makes it different
Most agent frameworks give you one loop: the model picks a tool, calls it, reads the result, picks the next. If you want to change course, you cancel and start over.
Unity gives every operation its own loop and returns a **live handle** you can steer. These handles nest — the user steers the ConversationManager, the ConversationManager steers the Actor, the Actor steers the managers. Corrections, pauses, and queries propagate through the full depth.
In practice:
* "Also include Q2 numbers" mid-way through a report → the agent adjusts without restarting
* "Pause that, something urgent" → work freezes and resumes exactly where it left off
* "How's the flight search going?" → status update without disrupting the work
* Three tasks running at once, each independently steerable
## Architecture at a glance
```
ConversationManager (dual-brain orchestration)
│
│ Slow Brain ◄── IPC ──► Fast Brain (real-time voice)
│
▼
CodeActActor (writes Python plans over typed primitives)
│
▼
State Managers (each runs its own async LLM tool loop)
├── ContactManager — people and relationships
├── KnowledgeManager — domain facts
├── TaskScheduler — durable tasks and execution
├── TranscriptManager — conversation history
├── GuidanceManager — procedures and SOPs
├── FileManager — file parsing
├── WebSearcher — web research
├── SecretManager — encrypted secrets
└── ...more
│
├── EventBus — typed pub/sub backbone
└── MemoryManager — offline knowledge consolidation
```
## Deep dives
The universal protocol for mid-flight control. Pause, resume, interject, and query any running operation at any depth.
The agent writes Python programs over typed primitives — not flat JSON tool calls. Real variables, loops, and control flow.
A slow deliberation brain and a fast real-time voice agent, communicating over IPC. The agent keeps talking while working.
Distributed domain managers communicating through English-language APIs. Each runs its own LLM tool loop.
Every 50 messages, conversations are consolidated into structured, queryable tables — contacts, knowledge, tasks, policies.
Real LLMs, cached responses, parallel execution. Never mocked.
## Get started
→ [Quickstart](/basics/quickstart) — clone, install, run the sandbox in 5 minutes
All three repos ([unity](https://github.com/unifyai/unity), [unillm](https://github.com/unifyai/unillm), [unify](https://github.com/unifyai/unify)) are MIT-licensed. The full product — with voice calls, messaging channels, and a management dashboard — runs on [Unify's platform](https://unify.ai).