Theo Memory
Theo Memory is the personal knowledge graph that remembers your facts, preferences, and instructions across conversations — so Theo gets more useful the more you use it.
In depth
Memory is split into multiple types — facts (people, projects, products), preferences (tone, format, style), instructions (always include X), and entities derived from the things you build. Memories are extracted automatically by Theo and can also be saved explicitly.
Retrieval is tiered: fast keyword retrieval for exact matches, plus an Upstash Vector index for semantic recall. The MCIR pipeline (Memory-Completion Intent Resolution) blends memory hits into intent classification so Theo can resolve “update the deck” to the right project.
You can review, edit, and delete every memory from settings. Memory is private to your account and never leaves your tenant.
Examples
- “My brand color is #FF5C00” persisted across decks
- “Always write in Spanish unless I ask otherwise”
- “Project Aeternum = the SaaS rewrite I’m running”
Also known as
Want to put this concept to work in OpenCharts?
How memory worksRelated terms
MCIR (Memory-Completion Intent Resolution)
MCIR is the intent-resolution pipeline that combines fast regex routing, a small reasoning model, behavioral patterns, and memory context to decide which Theo mode and tools should handle each message.
Intent Classifier
The Intent Classifier is the layer that decides which Theo mode a message belongs to — fast, think, code, image, video, social, learn, research, or others — using regex patterns plus a small AI scorer.
Theo Personality
Theo Personality is your per-account configuration — tone, persona, language, formatting preferences — that gets injected into every system prompt so Theo always sounds like the assistant you want.