One project. One method.
A team of Agents working together.
Axis doesn't give you a chatbot — it gives you a team of agents. Every agent has a role, but every agent coordinates around the same method and the same files. A Main Agent orchestrates — everyone else leaves an audit trail that can be reviewed, evaluated and potentially reversed.
01/ Team
Main Agent
"Axel" is your project coordinator. It boots the session, plans the work, delegates tasks, keeps records, and reports back to you. There is one Main — and only one Main — per project. Axis ensures agents don't work at cross-purposes.
Subagents
Subagent specialists handle specific jobs with a smaller and fresh context: research, analysis, writing, cross-checking. Instructions come in; verified results come out — nothing is left running.
Cross-Examiner
An independent skeptic stress-tests important work — all of the assumptions, logic, and evidence behind each step. It records its findings and pushes back on the work of other agents, whether that is flattering or not.
Librarian
A Wiki agent ingests sources and builds a cross-linked knowledge graph of pages, citations, and other resources. Knowledge compounds and improves over time, and everything happens automatically behind the scenes.
Intern
Local models can provide free labor for bounded, mechanical work. After checking the capabilities of your machine, Axis manages everything behind the scenes with Ollama.
Assistant
An external agent can run beside the Main Agent for read-only or append-only work. With optional integration with OpenClaw, you can reach your assistant by WhatsApp, email, or other channels for remote work.
02/ Rules
One project — one Main.
A second agent that shows up while your session is live stands down to a bounded role — it never silently becomes a second driver. Handover is an explicit, disclosed, user-confirmed ceremony.
Multiple agents — one method.
Many agents can serve one project without working at cross-purposes: every agent follows the same method, proves it is authorized before touching shared files, and stops cleanly when its job is done — or when you say stop.
Audit history — not hallucination.
Logs, snapshots, and reports are never edited after the fact. What the agents did, decided, and got wrong stays on file — that's what makes an audit worth anything.
Secure sources — not backdoors.
A document that tries to give your agents instructions gets quoted as a finding, not obeyed. Delegation briefs are sealed at both ends so a truncated brief fails loudly instead of working quietly.
03/ Rigor
Startup that cannot be skipped.
Every session begins the same way: the agent reads the protocol, checks who else is live, and takes its lease before it says a word. There is no "quick answer" path around it — the discipline runs before the first reply.
Records that write themselves.
Logs, snapshots, tasks, follow-ups, and reminders are written as a side effect of the work, not as a chore afterwards. Nobody has to remember to keep the journal, so the journal is always current.
Verification before delivery.
Agents check their own work against the requirements, cross-examine what matters, and say plainly what was verified and what was not. Confidence is earned, not asserted.
Rules that outrank convenience.
Practices are loaded when they are needed and never worked from memory. Small models get bounded jobs; only a capable model drives. A shortcut fails loudly instead of quietly.
Workflows fail when discipline stops.
Axis doesn't run on willpower. Discipline lives in the folder, not in anyone's memory: every agent, every session follows the same method, keeps the same records, and proves its work. You get discipline without practicing discipline.