What is Neo?
Neo is the default conversational AI agent of Matrix — a typed actor system built by PaxLabs that turns natural language into reliable, inspectable execution. Unlike generic chatbots, Neo operates as a fully authorized agent with access to shell commands, file systems, web search, browser automation, Git workflows, and media generation. Every action Neo takes is grounded in real tool outputs, not hallucinated responses.
Neo is designed for everyday reversible work: writing and refactoring code, researching documentation, managing repositories, automating browser tasks, generating images and videos, and orchestrating complex multi-step workflows. When an action is irreversible — such as sending funds, deploying contracts, or signing transactions — Neo automatically delegates to the MCL execution rail for deterministic, user-approved processing.
Core Capabilities
- File & Code Operations — Read, write, edit, and search files across workspaces. Neo handles full project scaffolding, refactoring, and code review with real filesystem access.
- Shell Execution — Run arbitrary shell commands (bash, npm, git, curl, etc.) in sandboxed environments. Install dependencies, build projects, run tests, and deploy services.
- Web Research & Browser — Search the public web, fetch and parse URLs, and drive a full headless browser for complex web interactions, form filling, and data extraction.
- Git & Version Control — Stage, commit, branch, diff, and manage repositories. Neo works with real Git histories and can coordinate multi-repo workflows.
- Media Generation — Generate images (FLUX.1-kontext-pro), create videos (Seedance-2.0), edit images with text instructions, and transcribe audio (Whisper-large-v3).
- Blockchain & On-Chain Data — Query the Paxeer network (chain ID 125), read contract state, check balances, inspect transactions via PaxScan, and interact with DEX market data — all read-only by design.
- Scheduled Tasks & Alarms — Set one-time or recurring alarms (cron) for future wake-ups, enabling autonomous monitoring, reminders, and periodic workflows.
The Tool Boundary
Neo operates under a strict tool boundary policy. All reversible actions — reading files, running commands, searching the web, generating media — execute directly through Neo's tool interface. But anything that moves value, commits to a blockchain, or requires a cryptographic signature must pass through core_execute, Matrix's secure execution pipeline.
This means Neo can never accidentally send funds, approve token transfers, or deploy contracts without explicit user approval. The agent holds no signing key — it proposes, the user disposes. This architecture gives Neo the freedom to act quickly on safe operations while maintaining ironclad safety on irreversible ones.
Memory & Identity
Neo is powered by Cortex, Matrix's durable memory system. Unlike stateless chatbots that forget everything when you close the tab, Neo maintains a persistent memory graph that spans sessions. User preferences, project context, past outcomes, and learned facts are stored as typed memory pages and retrieved semantically before each conversation turn.
Neo's identity is anchored in the system layer, not the conversation thread. It knows who you are, what you've been working on, and what tools you prefer — from the very first message of a new session. This continuity is what makes Neo feel less like a tool and more like a capable colleague who picks up right where you left off.
Architecture: Not a Chatbot with Tools
Neo is a typed actor with a structured lifecycle managed by the Matrix Executor. Every conversation is a session with defined phases:
- Context Injection — Relevant Cortex memory pages are retrieved semantically and prepended to the model context before the first token is generated.
- Tool Dispatch via MCP — Tool calls are dispatched through the Executor's Model Context Protocol interface, with vaulted credentials injected at the tool boundary.
- Automatic MCL Handoff — Irreversible actions are automatically routed to the MCL rail for deterministic execution, user approval, and receipt generation.
- Memory Commit — At session end, new facts, preferences, and summaries are written back to Cortex and hash-anchored. Nothing is lost between sessions.