How the Brain Works
A local-first vault, 83 skills, 7 agents, and a governance system that keeps you in control.
The Architecture
Agent0 has four layers. Each one builds on the last:
| Layer | What It Does | Built On |
|---|---|---|
| The Vault | Stores everything. Files, notes, captures, artifacts. | Obsidian |
| Skills | Define what the system can do. 83 reusable instructions. | Claude Code |
| Agents | Decide when and how to act. 7 specialized team members. | Claude Code |
| Governance | Keeps you in control. Approval matrix and escalation rules. | Vault files |
Everything runs locally. Your vault is an Obsidian workspace on your machine, connected to Claude Code for AI capabilities. Files sync to GitHub for backup and cross-device access. No cloud database. No vendor lock-in. You own every file.
The Vault
The vault is an Obsidian workspace with a specific folder structure. Each folder has a clear purpose:
| Folder | Purpose |
|---|---|
| daily/ | Universal entry point. Brain dumps, captures, agent standups. The daily note IS the inbox. |
| ideas/ | Business concepts and app ideas. Only one idea can be active at a time. |
| tasks/ | Folder-based pipeline: backlog, ready-to-start, in-progress, waiting-for-review, done. |
| projects/ | Active work with status and next actions. |
| decisions/ | What was decided, why, and what changed. Your future self will thank you. |
| artifacts/ | Everything the system produces. Briefs, articles, page copy, plans. |
| knowledge-base/ | Reference material. TikTok digests, YouTube summaries, repo analyses. |
| personal/ | Life outside work. Family, investments, health, goals. |
| radar/ | External market signals. Competitor moves, pricing changes, emerging trends. |
Every folder has a hub file that explains what belongs there, how content flows in and out, and which skills and agents touch it. The vault is self-documenting.
Skills
Skills are reusable instructions that tell the AI exactly how to perform a task. You invoke them with slash commands:
/dailyto start your journal/captureto process a link or video/get-shit-doneto execute approved tasks/distributeto create platform-native content/reflectto generate weekly and monthly reflections
There are 83 skills across 10 categories:
| Category | Count | Examples |
|---|---|---|
| Workflows | 14 | daily, capture, sync, tldr, publish |
| Marketing | 27 | copywriting, page-cro, seo-audit, content-strategy |
| Thinking | 6 | trace, connect, challenge, drift |
| Tools | 10 | obsidian-markdown, vault-setup, defuddle |
| Meta | 7 | brainstorming, writing-plans, skill-evolution |
| + 5 more | 19 | Sales, Dev, Reviews, Advertising, Services |
Skills load on demand. Only 3 to 5 are active in any session based on what you are doing. This keeps the system fast and focused.
Agents
Seven specialized agents handle different domains. Each has its own personality, scoped skills, and rules about what it can and cannot do.
| Agent | Role | What They Handle |
|---|---|---|
| Cortex | Chief of Staff | Coordination, task routing, scheduling. Runs interactive and unattended. |
| Loom | Creative Director | Copy, content strategy, brand voice. |
| Radar | Head of Growth | Market signals, competitors, conversion optimization. |
| Hippocampus | Head of Research | Knowledge capture, link processing, SEO. |
| Signal | Social Media Director | Platform-native content, distribution, scheduling. |
| Sentinel | Operations Analyst | Vault health, broken links, service monitoring. |
| Axon | Code Reviewer | Implementation review, code quality, verification. |
Two agents also run on a schedule without you. Cortex handles inbound work: task proposals, nightly journal entries, reflections. Radar handles outbound work: repo monitoring and signal detection.
The Governance Layer
The governance system is what makes this safe. Every action is classified:
| Classification | Examples |
|---|---|
| Auto-execute | Daily note processing, nightly journal entries, reflections, content drafts |
| Requires approval | Task execution, publishing to social media, skill modifications, external API calls |
Four governance documents define the boundaries:
- Approval Matrix defines what each agent can auto-execute vs. what needs your sign-off
- Agent Permissions sets per-agent access and capability limits
- Escalation Rules define when agents must stop and ask you
- Rejection Log tracks patterns so agents learn what you do not want
How It All Connects
Here is the flow from brain dump to shipped output:
- You brain dump into the daily note
- The system extracts tasks, routes ideas, logs decisions, processes captures
- Agents analyze what you wrote and connect it to existing vault knowledge
- Tasks flow through the pipeline (backlog, approved, in-progress, done)
- Content gets drafted, reviewed, and published
- Everything syncs via git across all your devices
The vault is the memory. Skills are the capabilities. Agents are the team. Governance is the rulebook. Together, they form a second brain that actually works.