secbrain 0.1.0

Persistent memory for Claude Code — local or VPS

SecBrain

Persistent memory for Claude Code — local or VPS.

SecBrain gives Claude Code a long-term memory vault that persists across sessions. Install it once globally, run secbrain init in any project, and Claude will automatically remember bugs, features, decisions, and learned patterns.

---

Install

pip install secbrain

Quick Start

# In any project directory
cd my-project
secbrain init

The wizard will:

1. Ask where your vault lives (local ~/.secbrain/vault or remote VPS)
2. Deploy the MCP server (locally or via SSH)
3. Register the current project
4. Inject CLAUDE.md instructions and .mcp.json MCP config
5. Install Claude Code session hooks

Then start a Claude Code session — memory is active automatically.

---

Commands

Command	Description
secbrain init	Full setup wizard
secbrain connect [name]	Connect current directory to a vault project
secbrain projects	List all vault projects
secbrain status	Show server health and current project
secbrain start	Start local vault server
secbrain stop	Stop local vault server
secbrain logs [project]	Show session log
secbrain bugs [project]	Show known bugs
secbrain deploy	(Re)deploy server to VPS
secbrain eject	Remove SecBrain from current project
secbrain skills	List learned skills
secbrain skills search <query>	Search skills
secbrain skills show <name>	Show a skill

---

What Gets Stored

~/.secbrain/vault/
├── _system/
│   ├── log.md          ← all session start/end events
│   └── entities.db     ← SQLite knowledge graph
├── projects/{name}/
│   ├── overview.md     ← project summary
│   ├── bugs.md         ← logged bugs
│   ├── features.md     ← logged features
│   ├── decisions.md    ← architectural decisions
│   ├── sessions.md     ← session summaries
│   └── user_model.md   ← learned user preferences
├── skills/
│   ├── global/         ← cross-project reusable patterns
│   └── {project}/      ← project-specific patterns
├── raw/                ← knowledge ingestion staging
└── wiki/               ← processed topic wikis

---

MCP Tools Available to Claude

Tool	When to call
vault_start_session(project, cwd)	Automatically via hook at session start
vault_end_session(project, summary)	At end of session
vault_log_bug(project, title, description, severity)	When discovering a bug
vault_log_feature(project, title, description, status)	When implementing a feature
vault_log_decision(project, title, rationale)	When making an arch decision
vault_recall(project, query)	Search everything
vault_update_overview(project, overview)	Update project description
vault_save_skill(...)	After solving something non-obvious
vault_recall_skill(query, project)	Before starting complex tasks
vault_improve_skill(project, skill_name, refinement)	When a skill gets better
vault_update_user_model(project, observation)	When noticing user patterns
vault_read(path)	Read any vault file
vault_write(path, content)	Write any vault file
vault_list(path)	List vault directory
vault_list_projects()	List all projects

---

VPS Deploy

secbrain init
# Select "VPS" → enter user@host → confirm deploy

SecBrain installs itself on the VPS via pip and runs as a systemd service on port 8765.

---

Multiple Projects

cd ~/other-project
secbrain connect
# Select existing project or create new

---

Development

git clone ...
cd secbrain
pip install -e ".[dev]"

---

Requirements

• Python 3.10+
• Claude Code with MCP support
• For VPS: SSH access + pip on remote

---

License

MIT