The Problem With AI Assistants

You tell ChatGPT about Sarah on Monday. By Wednesday, it's forgotten.

You make a promise in a meeting. Nobody tracks it. You promise a deliverable on Friday and lose track of it by Wednesday. You meet someone important, and three weeks later you can't remember what you talked about.

AI tools don't have memory. Claudia does.

What Makes Claudia Different

How Her Mind Works

Quick Start

1. Install

npx get-claudia

2. Start

cd claudia
claude

3. Say hi. She'll introduce herself, learn about you through a natural conversation, and generate a personalized workspace.

What's next:

• /morning-brief to see what needs attention
• Tell her about a person and she'll create a relationship file
• Share meeting notes and she'll extract action items
• npx get-claudia google to connect Gmail, Calendar, Drive, and more

Requirements: Claude Code, Node.js 18+, Python 3.10+ (for memory), Ollama (for embeddings)

Embeddings model: After installing Ollama, pull the required model: ollama pull all-minilm:l6-v2

npx get-claudia my-project --no-memory

Installs only the template layer (skills, commands, rules). Claudia works using markdown files without semantic search or pattern detection. You can add the memory system later by running the installer again without --no-memory.

cd your-claudia-directory
npx get-claudia .

This upgrades framework files (CLAUDE.md, skills, rules, daemon code) while preserving your data (context/, people/, projects/, databases). Safe to run multiple times. The installer detects existing installations and only updates system files.

See It in Action

You:     What do I know about Sarah Chen?

Claudia: Sarah Chen is CEO at Meridian Ventures. You met at SaaStr 2025.

         📊 Relationship Health: Strong (last contact 3 days ago)

         🎯 Open Commitments:
         • Send updated pitch deck to Sarah [OVERDUE - was due Friday]

         💡 Recent Context:
         • She's interested in AI infrastructure angle
         • Wants to see technical architecture doc
         • Bringing in technical partner for next meeting

         🔗 Connected to: Series A (project), Priya Sharma (introduced by)

         Source: Call notes from Jan 28, Email thread Jan 30

Morning brief pulls everything together:

You:     /morning-brief

Claudia: ☀️ Morning Brief — Wednesday, Feb 19

         ⚠️ Needs Attention:
         • Send updated pitch deck to Sarah Chen [OVERDUE]
         • Quarterly review with Acme Corp is tomorrow

         👀 Watch:
         • Haven't spoken to Marcus in 18 days (usually weekly)
         • Two new commitments this week, zero closed

         📊 Today: 3 meetings, 2 open commitments

Judgment rules shape everything she does:

You:     I've got the Acme proposal and Marcus wants a catch-up.
         Which first?

Claudia: Acme proposal. Your judgment rules say revenue-generating
         work beats relationship maintenance when there's a
         deadline, and the Acme proposal is due Thursday.

         I'll remind you about Marcus tomorrow morning.

Pattern detection works across weeks, not just moments:

Claudia: I've noticed something across the last few weeks.
         You've taken on 4 new commitments without closing any.
         Last time this happened (early January), two slipped
         past deadline.

         Want me to flag the lowest-priority ones so you can
         decide what to defer?

Every conversation builds on the last. She remembers.

Who It's For

Claudia detects your work style and generates structure that fits:

| Role | What You Get | |------|--------------| | Consultant | Client folders, deliverable tracking, pipeline commands | | Executive | Direct reports, initiatives, board prep, leadership tools | | Founder | Investor CRM, fundraising cycles, team and product tracking | | Solo Professional | Flexible structure for wearing many hats | | Creator | Audience growth, content calendar, collaboration tracking |

Key Commands

| Command | What It Does | |---------|--------------| | /morning-brief | What needs attention today: commitments, meetings, warnings | | /new-workspace [name] | Spin up a new project workspace from templates | | /meeting-prep [person] | One-page briefing before a call | | /capture-meeting | Process notes into decisions, commitments, action items | | /what-am-i-missing | Surface risks, overdue items, cooling relationships | | /research [topic] | Deep research with web sources and memory integration | | /inbox-check | Lightweight inbox triage across connected email accounts | | /brain | Launch 3D brain visualizer | | /meditate | End-of-session reflection: extracts learnings, judgment, patterns | | /deep-context [topic] | Full-context deep analysis | | /memory-audit | See everything Claudia knows, with source chains |

| Command | What It Does | |---------|--------------| | /weekly-review | Guided reflection across relationships and projects | | /growth-check | Periodic reflection on development, skills, and progress | | /financial-snapshot | Revenue, expenses, invoicing, and cash flow metrics | | /draft-reply | Email response drafts in your voice | | /follow-up-draft [person] | Post-meeting thank-you and summary | | /new-person [name] | Create a relationship file | | /pipeline-review | Active deals, capacity, pipeline health | | /client-health | Status across all client relationships | | /databases | View, switch, and manage memory databases | | /brain-monitor | Launch the Brain Monitor TUI dashboard | | /fix-duplicates | Find and merge duplicate entities | | /memory-health | Check memory system health | | /diagnose | Check memory daemon health and troubleshoot |

Plus ~30 proactive skills (commitment detection, pattern recognition, judgment awareness, cognitive extraction, risk surfacing, and more) that activate automatically based on context.

Brain Visualizer

Launch with /brain to see your memory as a 3D network graph. Entities are nodes, relationships are edges, and everything is interactive: click to inspect, filter by type, search by name.

Integrations

Claudia works fully on her own, but integrations let her see further.

Google Workspace

Connect Gmail, Calendar, Drive, Docs, Sheets, Tasks, and more with a single setup command:

npx get-claudia google

This generates a one-click URL to enable all required Google APIs and walks you through OAuth setup. Three tiers available:

| Tier | Tools | What You Get | |------|-------|-------------| | Core | 43 | Gmail, Calendar, Drive, Contacts | | Extended | 83 | Core + Docs, Sheets, Tasks, Chat | | Complete | 111 | Extended + Slides, Forms, Apps Script |

500+ Apps via Rube

Rube (by Composio) connects Claudia to Slack, Notion, Jira, GitHub, Linear, HubSpot, Stripe, Figma, and hundreds more through one-click OAuth. No per-app MCP setup needed.

| Category | Examples | |----------|----------| | Communication | Slack, Discord, Teams, Telegram | | Project Management | Jira, Linear, Asana, Trello, Monday.com | | Knowledge & Docs | Notion, Confluence, Google Docs, Coda | | Code & Dev | GitHub, GitLab, Bitbucket | | CRM & Sales | HubSpot, Salesforce, Pipedrive | | And 500+ more | Browse the full list |

Obsidian Vault

Memory auto-syncs to an Obsidian vault at ~/.claudia/vault/ using PARA structure. Every entity becomes a markdown note with [[wikilinks]], so Obsidian's graph view maps your network. SQLite is the source of truth; the vault is a read-only projection you can browse and search.

How It Works

41 skills · 33 MCP tools · 500+ tests

Claudia has two layers:

Template layer (markdown) defines who she is. 41 skills, rules, and identity files that Claude reads on startup. Skills range from proactive behaviors (commitment detection, pattern recognition, judgment awareness) to user-invocable workflows (/morning-brief, /research, /meditate). Workspace templates let you spin up new projects with /new-workspace [name].

Memory system (Python) defines what she remembers. Two daemon modes share the same SQLite database:

| Daemon | When | Purpose | |--------|------|---------| | MCP daemon | Per-session (stdio) | Serves ~33 memory tools to Claude Code | | Standalone daemon | 24/7 (LaunchAgent) | Runs scheduled jobs even when Claude Code is closed |

| Scheduled Job | When | What It Does | |---------------|------|--------------| | Adaptive decay | 2 AM | Fades old memories, high-importance at half rate | | Consolidation | 3 AM | Merges duplicates, detects patterns, tracks relationships | | Vault sync | 3:15 AM | Syncs memory to Obsidian vault (PARA structure) | | Pattern detection | Every 6h | Surfaces trends across conversations |

You ──► Claude Code ──► Reads Claudia's templates ──► Becomes Claudia
                                                           │
                                                    MCP daemon (stdio)
                                                           │
                                                           ▼
                                                     SQLite + vectors
                                                           ▲
                                                           │
                                              Standalone daemon (24/7)
                                               ┌──────┼──────┐
                                               ▼      ▼      ▼
                                          Scheduler Ollama  Obsidian vault
                                                           (PARA structure)

Agent team for speed. Claudia delegates structured work to a two-tier team. Tier 1 (Haiku): fast workers for document archiving, processing, and schedule analysis. Tier 2 (Sonnet): a research scout with independent context for multi-turn web research. Claudia keeps relationship judgment and strategy decisions for herself.

Semantic search uses hybrid ranking: 50% vector similarity, 25% importance, 10% recency, 15% full-text. Accessing a memory boosts it (rehearsal effect).

Judgment layer stores user-defined decision rules in context/judgment.yaml. Rules are extracted during /meditate reflections and applied by morning briefs, commitment detection, and risk surfacing. Claudia learns your priorities once and applies them consistently.

Document storage keeps files, transcripts, emails on disk, linked to people and memories. Content-hash deduplication. Automatic lifecycle management.

Provenance chains trace any fact to its source email, transcript, or conversation.

Graph traversal connects dots across your network. Ask about one person, see related entities with top memories. The 3D brain visualizer (/brain) renders the graph in real-time.

Per-project isolation keeps work memories separate from personal. Each workspace gets its own database.

Session reflections (/meditate) extract learnings about your preferences, communication patterns, and judgment calls. These persist across sessions and make Claudia progressively sharper.

For full architecture diagrams, see ARCHITECTURE.md.

Privacy and Safety

• Fully local. Memory, embeddings, cognitive tools run on your machine. No external APIs for storage.
• No external actions without approval. Every email, calendar event, external action requires your explicit "yes."
• Your data in two formats. SQLite database (~/.claudia/memory/) for fast semantic search, plus a PARA-organized Obsidian vault for reading and graph navigation. Two independent copies you own forever.
• Delete anything, anytime. Full control over your data. No lock-in, no cloud dependency.

System Requirements

| Component | Required | Purpose | |-----------|----------|---------| | Claude Code | Yes | Runtime | | Node.js 18+ | Yes | Installer | | Python 3.10-3.13 | Recommended | Memory system | | Ollama | Recommended | Embeddings + cognitive tools |

Without the memory system, Claudia still works using markdown files. With it, she gains semantic search, pattern detection, and relationship tracking.

Ollama model: Run ollama pull all-minilm:l6-v2 after installing Ollama. This is the embedding model used for semantic search.

Platforms: macOS, Linux, Windows

Memory tools not appearing in Claude Code?

1. Check .mcp.json has a claudia-memory entry with the correct venv Python path
2. Restart Claude Code in a NEW terminal
3. Re-run the installer to fix paths: npx get-claudia .

Check standalone daemon health:

curl http://localhost:3848/status
launchctl list | grep claudia
tail -20 ~/.claudia/daemon-stderr.log

Standalone daemon not running?

launchctl load ~/Library/LaunchAgents/com.claudia.memory.plist

Ollama not running after reboot?

open -a Ollama          # macOS
ollama serve            # Linux

Pull models manually:

ollama pull all-minilm:l6-v2    # Embeddings (required)

Google Workspace not working after enabling new APIs? Delete the cached token and restart to re-authenticate with updated scopes:

rm ~/.workspace-mcp/token.json
# Restart Claude Code

Broken install? Re-run setup:

cd your-claudia-directory
npx get-claudia .

This updates daemon code, skills, and rules while preserving your databases and context files.

Contributing

Claudia is source-available under the PolyForm Noncommercial License 1.0.0.

• Template (skills, rules, identity): template-v2/
• Memory daemon (Python): memory-daemon/ (tests: cd memory-daemon && pytest tests/)
• Installer: bin/index.js
• Architecture: ARCHITECTURE.md
• Dev guide: CLAUDE.md

License

PolyForm Noncommercial 1.0.0

Free for personal, research, educational, and nonprofit use. Commercial licensing: [email protected]