The Axis agent avatar - a friendly robot
The agents

One project. Many agents.
One integrated workflow.

The Axis Workflow doesn't give you a chatbot. It gives you a cast of agents with defined roles - a project coordinator, independent specialists, a cross-examiner, a librarian, an assistant - even a cheap local intern for cheap labor - all working from the same system of plain markdown files - all leaving an auditable record.

/ The cast

Each role exists because a real failure mode exists without it.

Main Agent
Your project coordinator

The Main Agent starts the session, plans the work, tracks the tasks, keeps the records, and directs everyone else. You talk to it; it answers for the project. In Axis it even has a name: Axel.

Subagents
Your project specialists

The Main Agent spawns helpers for particular jobs with a fresh context: research a question, ingest a document into your knowledge base, run a routine transform on a cheap local model. They do the job, report back, and disappear.

Cross-Examiner
Your devil's advocate

Important work shouldn't grade itself. An independent agent with no stake in the answer stress-tests assumptions, logic, and output - and files its findings on the record, whether flattering or not.

External Agents
Your personal assistant

An always-on agent arriving beside a live session - say, from your phone - does not become a second driver. It reads everything, answers anything, takes notes, and drafts new documents with a provenance stamp - but cannot edit a single existing file or touch a secret.

/ Your agents build you a knowledge base

Drop source documents into the Wiki inbox and a librarian subagent reads, summarizes, and cross-links them - a knowledge graph you own, in plain markdown, readable by any tool.

An Axis knowledge wiki in Obsidian's graph view - pages written and cross-linked by the agents

A real Axis wiki in Obsidian's graph view - purple syntheses, green concepts, orange entities, yellow deals - written and linked entirely by the agents.

/ Tested infrastructure

The agent roles above are not aspirations - they are rehearsed, live, until failure is boring.

Every release passes a suite of 89 automated integrity checks, run on macOS and Linux on every push - and that mechanical suite is the floor, not the story. The guarantees are rehearsed on real hosts: real messaging channels, real cron jobs, real concurrent sessions racing for the same identifiers. We boot cold agents beside live ones and verify the newcomer stands down. We plant kill signals between a request and its write and verify the dying session refuses to write. We truncate delegation briefs at both ends and verify the fragment fails loudly. When a drill fails, the failure becomes doctrine, the doctrine becomes a check, and the failing scenario is re-run until it passes. We publish numbers next to claims, and we say tested - never guaranteed - because knowing the difference is part of the rigor.

“A discipline layer for AI-assisted knowledge work… It gives an LLM project what software engineering already has: durable state, an audit trail, session continuity, independent review, and portability. The pieces reference each other consistently… The design holds together.”

— Claude Fable 5 (Anthropic), asked for an independent technical assessment of the Axis Workflow

Meet your own cast.

Get the Axis Workflow free, or let SimAxis onboard your team - first project, first delegation, first cross-examination, on your real work.