What HiveXS is.

A map of the components that make up HiveXS, how they connect, and what each one does. The main agent is the front door. Everything else supports it.

HiveXS started as a way to close the gap between having an idea and doing something about it. Most agency work loses time in the middle — between knowing what to do and actually getting it done. This is the system that closes that gap.

At its core it is a set of agents that can read, think, write, and act. They have access to your workspace, your email, your files, and a growing set of cloud tools and services. You brief them like a person. They run the job, produce the output, and report back. The more you use it, the more it knows.

INTERNAL ARCHITECTURE SCHEDULED JOBS Email watcher every 2 min Knowledge review nightly 23:30 Knowledge backup nightly 03:00 Cost reconciler nightly 23:25 Weekly review Mon 09:00 Main Agent HiveXS reads, routes, responds KNOWLEDGE Knowledge Layer SQLite + BGE-small embeds Llama 3.2 re-ranker indexes the workspace Cost Ledger ~/.openclaw/cost_ledger.db CHANNELS Email claw01@hive.au Slack #hivexs-agents Webchat OpenClaw interface WORKSPACE Drive Folders 01 Inbox · 02 Active 03 Output · 04 Archive 00 System SPECIALIST Strategist thinking, planning, frameworks, strategy LOCAL MODELS Ollama Service Gemma 3 27B Llama 3.2 (re-ranker) Phase A done, Phase B live Future agents writer, researcher, content inbound dispatch read / write API calls re-ranking indexed triggers review, backup reconciler
main agent dispatches to cloud models and service APIs
EXTERNAL CONNECTIONS CLOUD AI MODELS SERVICE APIS Anthropic Claude Opus / Sonnet / Haiku current default model OpenAI GPT class + image models connection available Google AI / Gemini Nano Banana image model connection available HubSpot CMS + Files content pipeline SHWR + future client work Freepik API image / asset library visual resources on demand Google Workspace Gmail / Drive / Docs / Sheets claw01@hive.au OAuth live
Main agent (front door)
System component
Data or dispatch connection
Scheduled trigger
External connection

Main Agent

The front door. Everything that comes in passes through here first. Reads emails, reads inbox files, decides what to do, responds, and dispatches to specialist agents when the job calls for it.

Strategist

A specialist subagent focused on thinking and planning work. Dispatched by the main agent when a job needs structured strategic output. Writes directly to 03 Output and reports back.

Channels

The three ways people reach HiveXS. Email to claw01@hive.au, Slack in #hivexs-agents, or webchat via the OpenClaw interface. All inbound traffic surfaces to the main agent.

Knowledge Layer

A local search index built from the workspace folders. Uses SQLite storage, BGE-small embeddings to encode content, and a Llama 3.2 re-ranker to surface the most relevant results. It knows what we've worked on.

Workspace Folders

The five Drive folders that structure all work. 01 Inbox for new jobs, 02 Active for work in progress, 03 Output for finished deliverables, 04 Archive for completed work, 00 System for agent config and memory.

Local Models

Ollama running two models locally. Gemma 3 27B for general reasoning, Llama 3.2 for re-ranking knowledge search results. Phase A (Gemma) done. Phase B (re-ranker) live.

Scheduled Jobs

Five cron jobs that run automatically. Email watcher every 2 minutes, knowledge review and backup nightly, cost reconciler at 23:25, weekly review Monday 09:00. They keep the system current without anyone having to ask.

Cost Ledger

Every API call is attributed and written to a local SQLite database. The nightly reconciler tallies usage and emails Wayne a summary. Nothing runs without a paper trail.