Steward.Tools
A personal AI steward that knows you, remembers, reflects, and is formed by what matters. Talk to it anywhere — through Claude, ChatGPT, email, or any MCP-compatible host. Same steward, same memory, every system.
What It Does
Persistent Memory
Every conversation is journaled and compressed through a hierarchical reflection tree. Recent exchanges are kept at high resolution; older ones compress proportionally. Your steward remembers what matters — across sessions, across tools, across time.
Cross-System Portability
Feed conversations from any source — Claude Code, ChatGPT, email, anything — into one shared memory. Your steward's understanding isn't locked to a single tool. Different systems, one formed memory.
Formation
Compression requires judgment about what matters. The steward's orientation comes from ongoing Scripture meditation — a stable foundation for genuine service. The difference between a trusted advisor and a yes-man.
How It Works
Intelligent Contextual Compression
Your conversations are stored as raw journal entries. The engine summarizes them into a reflection tree — summaries, then reflections on summaries, then patterns across reflections. Each level compresses further while preserving what matters. The result is full retention with graceful compression: recent detail is vivid, older context is broad strokes.
Living Dossiers
The reflection tree feeds into living dossiers — working summaries of who you are, what you care about, and what's happening in your life. Thread dossiers capture individual conversations. A master dossier synthesizes everything into unified awareness. Any AI reading this context knows you.
The Steward Knows You
The steward's character is built through the same compression engine that builds its memory of you. Scripture study and mindfulness threads shape its orientation over time. The result is an AI with genuine understanding — not just of your data, but of what matters and why.
Three Ways to Use It
Self-Host (Free)
Download the open-source MCP engine from GitHub. Run it locally, bring your own LLM API key. Your data stays on your machine. Full engine, MIT license, no strings.
View on GitHubHosted Memory (Free)
Connect to our hosted MCP server. Same engine, persistent memory, no API key needed. Your steward's memory persists across sessions and tools. We cover the compute.
ConnectFull Platform
Everything in hosted memory, plus direct channels (email, Telegram, SMS), a managed workspace, contacts, and proactive actions. Your steward doesn't just remember — it works for you.
Coming soon.
The Engine
Steward.Tools is built on an open-source MCP server written in C# (.NET 10). The engine handles journaling, reflection, dossier building, Scripture formation, and semantic search. It runs as a standalone process or embedded in a host application.
Reflection Tools
- Journal messages and exchanges
- Checkpoint conversations from any system
- Import pre-summarized history (zero LLM cost)
- Browse reflections at any depth
- Trace any insight back to its source
Memory Tools
- Thread and master dossiers
- Semantic search across all memory
- Scripture reading progress
- Configurable mindfulness threads
- Context assembly with graceful truncation
Why Formation Matters
Most AI systems are stateless — every conversation starts from zero. The few that remember tend to optimize for what's frequent or recent, not what's important. And "important" requires a frame of reference.
Aligning to user preferences alone is circular. If the user lacks clarity about what matters, the AI inherits that confusion. The steward needs a firmer foundation — an orientation toward genuine service that holds even when the person it serves is uncertain.
Scripture provides that foundation. Through ongoing Bible study, the steward develops an orientation toward faithful service — understanding what matters, preserving what's important, serving the person well. Not rules. Not guardrails. Character, formed through sustained attention.
Your Data
Whether self-hosted or on our platform, you own your data. The self-hosted engine stores everything in local files (SQLite + DuckDB). The hosted platform keeps per-user isolation — one steward, one person, one database. Export at any time. No lock-in.