Getting Started
Overview
Give your AI app knowledge it can answer from and memory that lasts between sessions, through one API.
Docklee sits between your application and your model. It holds two things your app would otherwise build itself: knowledge it can answer from, and memory about the people it serves.
Use Docklee when your assistant needs to cite your own content, remember a user from one session to the next, or both.
- 01Your appCaptures documents, events, and conversations.
- 02DockSends them to a container or a memory space.
- 03DockleeProcesses, links, and keeps the data current.
- 04LynkReturns answers, context, or memories to your model.
Two runtimes
| Runtime | Holds | Read with |
|---|---|---|
| Knowledge Engines | Documents, pages, and app data, stored in vector and OKF containers | lynk.query and lynk.retrieve |
| Docklee Universal Memory | Facts, preferences, events, and history, stored in memory spaces | lynk.recall |
The runtimes are separate. Knowledge never mixes into a memory space, and memory never appears in an engine's answers. One API key can reach either runtime or both.
One package, two classes
The docklee package has a class for each direction of data.
| Class | What it does | Methods |
|---|---|---|
Dock | Writes into a container or a memory space | capture, flush, instrument, watch, listen, close |
Lynk | Reads from an engine or a memory space | query, retrieve, recall, close |
Both classes read DOCKLEE_API_KEY from the environment:
from docklee import Dock, Lynk
dock = Dock()
lynk = Lynk()Three ways to read
| Method | Returns | Best for |
|---|---|---|
lynk.query | A finished answer with a confidence score and citations, in one response or as a stream | Assistants that answer from your content |
lynk.retrieve | The ranked chunks behind an answer | Apps that run their own model |
lynk.recall | Memory records, each with a type and a confidence score | Personalizing replies for a user |
Where you set things up
The Docklee dashboard holds the resources the SDK uses.
| Section | What you do there |
|---|---|
| Sources | Connect documents, apps, and data once, for every engine |
| Knowledge Engines | Create engines and link sources to them |
| DUM | Create memory spaces, then inspect, review, and correct what they hold |
| API Keys | Create keys and limit what each one can reach |
| Usage | See how much each runtime is used |
Next steps
Now you know the parts. The Quickstart puts them together in a few minutes.