# Previously

> an open-source personal AI agent that organizes memory by time - a single timeline of episodic slices, not chat threads

Previously is an open-source personal AI agent that organizes memory by time — a single timeline of episodic slices, not chat threads. This documentation covers concepts, guides, and reference material.

## Docs

- [Introduction](https://previously.ldwid.com/en/docs/introduction): What Previously is — a personal AI agent that organizes memory by time, forms a persistent understanding of you, and self-evolves after every conversation.
- [Why Previously](https://previously.ldwid.com/en/docs/why): Why time-based memory matters, why agents need a self-model that evolves, and how fixing the memory model fixes the interaction model.
- [The Timeline](https://previously.ldwid.com/en/docs/timeline): The timeline is Previously's core interaction model — a horizontal scrollable line where every dot is a conversation, from the past stretching to NOW. Click any dot and you're back in that moment.
- [Slices](https://previously.ldwid.com/en/docs/slices): How conversation bursts are captured as dual-timeline directories — core.md for what was said, agent.md for what the agent was thinking, plus a previously.md snapshot of its self-model at that moment.
- [Strands](https://previously.ldwid.com/en/docs/strands): How strands form the semantic keyword index across the timeline — the by-topic counterpart to the by-time slice record, now actively used by the recall system.
- [Memory Model](https://previously.ldwid.com/en/docs/memory-model): How Previously structures memory — the previously.md self-model with long-term and short-term memory, seven mutation operations, and pure-function evolution applied after every conversation.
- [Recall](https://previously.ldwid.com/en/docs/recall): How the agent searches your memory on demand — a Flash sub-agent scans the timeline and returns pointers, then the main model reads what it needs. The agent decides when to search, not the system.
- [Getting Started](https://previously.ldwid.com/en/docs/getting-started): How to use Previously — the two-screen layout, horizontal timeline, transparent thinking, self-evolution, and what day-to-day interaction looks like.
- [Deployment](https://previously.ldwid.com/en/docs/deployment): How to create a private copy, configure, and deploy Previously to Vercel in about 10 minutes — no server management or database setup required.
- [Configuration](https://previously.ldwid.com/en/docs/configuration): How Previously is configured — a capabilities module replaces manual env-var checks, reasoning effort is user-adjustable, and data-source auto-detection requires zero configuration for demo mode.
- [Architecture](https://previously.ldwid.com/en/docs/architecture): System architecture — two independent Vercel Workflows (chat + evolution) running in parallel, a Pull-based recall model where the agent decides when to search, and conceptual memory tools that replace filesystem operations.
- [FAQ](https://previously.ldwid.com/en/docs/faq): Frequently asked questions about Previously — how it works, what it costs, data privacy, and the architecture behind the self-model.

## Links

- GitHub: https://github.com/previously-lab/agent
- Live Demo: https://previously-demo.ldwid.com
- Dev.to Article: https://dev.to/likedreamwalker/is-time-the-missing-dimension-in-ai-memory-2l9c