Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.openori.dev/llms.txt

Use this file to discover all available pages before exploring further.

Memory is what separates OpenOri from every other AI app. While others start from zero every conversation, OpenOri remembers.

​
How it works

After every conversation, OpenOri runs a lightweight extraction pass that identifies:
  • Facts — “I work at Acme Corp”, “My project uses Next.js 15”
  • Preferences — “I prefer TypeScript strict mode”, “Use Tailwind v4”
  • Patterns — “When I ask about code, I usually mean the Atlas project”
  • Corrections — “Actually, I meant Python, not JavaScript” updates existing memory
This happens automatically. Just talk to OpenOri naturally — it learns on its own.

​
Example

Day 1:
You: “Help me set up a new Node.js project” OpenOri: Creates a basic JavaScript project
Day 30 (after learning your preferences):
You: “Help me set up a new Node.js project” OpenOri: Creates a TypeScript project with strict mode, ESLint, your preferred folder structure, and Drizzle ORM — because it remembers everything from previous conversations
This is the “gets smarter with use” effect. Every conversation teaches OpenOri something new.

​
The three tiers

OpenOri uses a tiered loading system to keep token usage efficient:
TierSizeWhat it holdsWhen loaded
L0 (Abstract)~100 tokensOne-line summary of each memoryEvery conversation
L1 (Overview)~2K tokensStructured detail, key factsWhen the topic is relevant
L2 (Full)VariableComplete context, full historiesOn explicit demand
OpenOri always has broad awareness of everything it knows (L0), loads details when needed (L1), and pulls full context only when deep-diving (L2).

​
Recall mode

The Recall toggle in the prompt box controls memory injection:
  • Recall ON — OpenOri searches its memory and includes relevant context in the conversation. It knows who you are and what you’re working on.
  • Recall OFF — Clean conversation with no memory. Useful for generic questions or when helping someone else.
Recall is ON by default. Toggle it off when you want OpenOri to respond without personal context.

​
Managing your memories

Open Settings → Context tab to:
  • View all memories — See everything OpenOri has learned, organized by category
  • Delete individual memories — Remove anything you don’t want remembered
  • View the Context Graph — A visual graph showing relationships between your memories, projects, and preferences
  • Check stats — Total memory count, project count, index size
OpenOri is fully transparent. You can see, edit, and delete everything it remembers. There are no hidden memories.

​
How memories are stored

~/.ori/context/
├── user/                    # About you
│   ├── brain.json           # Core identity: name, timezone, preferences
│   └── preferences/         # Domain-specific preferences
│       ├── coding.json      # "Uses TypeScript, prefers strict mode"
│       └── communication.json
├── projects/                # Your workspaces
│   └── <workspace-hash>/
│       └── context.json     # Tech stack, conventions, structure
└── agent/                   # Operational context
    └── memories/            # Extracted facts and patterns
Everything is plain JSON — portable and inspectable.