Core Concepts

The architectural reference for Axiom's behavior. Each page here explains how a part of the system works — the data model, the lifecycle, the contracts — rather than which button to click (that's Web UI) or which value to set (that's Settings).

Read these when you want to understand why the agent behaves the way it does, debug something that's not doing what you expected, or write a skill or instruction file that plays nicely with the existing machinery.

Pages

Page	What it covers
Agent Instructions	AGENTS.md, HEARTBEAT.md, CONSOLIDATION.md — the plain-Markdown files that shape day-to-day behavior.
Context Compaction	How long chats and tasks stay inside the model's context window: triggers, the summary format, budgets, /compact.
Built-in Tools	The tool registry: which tools the agent has every turn, which are opt-in, what each one does.
Codemode	The experimental scripted-tool-call tool: batching, chaining, filtering; the switches, the cronjob override, limits and the spill folder.
Multi-User	Isolation level and its limits, private vs. shared data, admin role, recommended setup.
Image Generation	Image models per provider, the default image model, the generate_image tool, output files, real cost tracking.
Memory System	The file-based memory tiers (SOUL.md, MEMORY.md, daily, users, wiki, sources) plus the SQLite fact store.
Skills	How skills are discovered, indexed, loaded on demand, and built-in vs. user-installed.
System Prompt	The full layered system prompt — every block the model sees before your message.
Tasks & Cronjobs	Background jobs, scheduled work, one-shot reminders — lifecycle, isolation, the scheduler.

Where to next?

• New to Axiom? Read Memory System and Skills first — they're the load-bearing concepts everything else builds on.
• Want to tune behavior? Concepts → Agent Instructions, then Settings for the knobs.
• Building a skill or task? Skills and Tasks & Cronjobs describe the contracts.