AI Process
User Guide — Seeing What the AI Has Learned and What It's Doing
| Last updated: September 2026
Contents
- Overview
- Key Concept: Knowledge vs Runtime
- Finding Your Way Around
- AI Knowledge
- AI Runtime
- Quick Reference
1. Overview
The AI Configuration screens (under Admin → AI) decide how AI is allowed to behave. AI Process is the other side of that coin: it's the operational, behind-the-scenes view of what the AI has actually done with your data — what it has learned, what it's still working through, and what it's doing for people right now.
It's split into two areas. AI Knowledge is about the data the AI has taken in and organised. AI Runtime is about the live activity — the chats it's answering and the suggestions it's making.
This is an administrative and monitoring area. Most of it is there to look at, check on, and occasionally nudge — not something you change day to day. It's where you go to confirm the AI is keeping up, spot anything stuck, and review the suggestions it produces.
WHERE TO FIND IT — AI Process is reached from the Process area (Process → AI Process). It has two pages, each with its own tabs: AI Knowledge and AI Runtime.
2. Key Concept: Knowledge vs Runtime
A quick way to keep the two pages straight:
| AI Knowledge | AI Runtime | |
|---|---|---|
| Answers the question | "What has the AI learned from our data?" | "What is the AI doing with it right now?" |
| Holds | Projects, embeddings, and what's searchable | Chat sessions, suggestions, and feedback |
| You mostly | Set up projects and check nothing is stuck | Review conversations and act on suggestions |
A word that comes up a lot here is embedding. An embedding is a numeric "fingerprint" of a piece of your data that lets the AI find things that mean something similar — it's how features like duplicate-matching and chat-style answers work. Turning your data into embeddings is the main job the Knowledge side is managing.
3. Finding Your Way Around
Go to Process → AI Process. You'll find the two AI Process pages there:
- AI Knowledge — four tabs: AI Projects, Embedding Store, Embedding Queue, Relevancy Index.
- AI Runtime — three tabs: RAG Sessions, Suggestions, Suggestion Feedback.
The rest of this guide walks through each tab in that order.
4. AI Knowledge
The AI Knowledge page is where the AI's understanding of your data is organised and kept up to date.
4.1 AI Projects
A project is a working context for the AI within a data domain — a named bundle that groups related knowledge together and ties it to a particular embedding model. The AI Projects tab lists them.
Adding a project
- Click + Add Project to open the Create AI Project window.
- Fill in the fields:
| Field | What it's for |
|---|---|
| Project Code | A short unique code for the project. |
| Project Name | A friendly name. |
| Description | What the project is for. |
| Project Type | The kind of work the project supports (for example, mapping). This decides how OnCoor finds the right project for a task, so choose it carefully. |
| Embedding Model | The model used to turn this project's data into embeddings. |
| Embedding Dimension | The size of those embeddings — set automatically to match the chosen model. |
- Save. The project appears in the list.
What each column shows: Org, Domain, Code, Name, Description, Type, Embedding Model, Dim (dimension), Status, Archived, and who last updated it and when.
Row actions: Edit a project's details, Archive one that's no longer active (and Reactivate it later), or Delete it.
PROJECT TYPE MATTERS — The Project Type is how OnCoor matches a task to the right project. Pick it from the list rather than typing a variation, so the AI can find the project when it needs it.
4.2 Embedding Store
The Embedding Store tab is the library of embeddings the AI has already created — one row per piece of data that's been turned into a fingerprint. You'd look here to confirm that a particular object has been taken in and is available to the AI.
Use the Project filter at the top to narrow the list to one project.
| Column | What it shows |
|---|---|
| Object Type / Object ID | What the embedding was made from, and its identifier. |
| Name | The name of the source object. |
| Source Table / Source Col | Where in your data the content came from. |
| Hash | A fingerprint of the exact content, used to avoid storing the same thing twice. |
| Content (preview) | A short preview of the text that was embedded. |
| Model / Dim | The model that made the embedding and its size. |
| Active | Whether this embedding is currently in use. |
| Indexed | When it was added. |
4.3 Embedding Queue
Turning data into embeddings happens in the background. The Embedding Queue tab is the to-do list for that work — each row is an item waiting to be processed, being processed, or recently finished. This is the tab to check if the AI seems to be missing recent data.
Filters: narrow by Org, Data Domain, Status (or All statuses), and a Stuck only switch to show just the items that need attention.
| Column | What it shows |
|---|---|
| Org / Domain | Who the item belongs to. |
| Object Type / Object ID | The item waiting to be processed. |
| Op | The operation to perform (for example, add or remove an embedding). |
| Status | Where the item is up to — for example waiting, in progress, done, or failed. |
| Lease | Whether a background worker currently has the item. An item marked STUCK has been held too long and needs a nudge. |
| Retries | How many times processing has been attempted. |
| Last Error | The message from the most recent failed attempt. |
| Queued / Processed | When the item joined the queue and when it finished. |
Row actions: Retry puts a stuck or failed item back to the start of the queue; Delete removes it.
WATCH FOR "STUCK" — Most items pass through the queue on their own. Turn on Stuck only now and then to check for items flagged STUCK; a Retry usually gets them moving again, and the Last Error tells you if something needs a closer look.
4.4 Relevancy Index
The Relevancy Index tab is a summary of what's actually searchable for the AI in each domain — a scoreboard showing, per scope, how much data is indexed and ready to be used in answers.
Filter by Data Domain, and use + Add Entry to add a scope to the index (or the Edit action to change one, or Delete to remove one).
| Column | What it shows |
|---|---|
| Scope | The area of data this entry covers. |
| Object Type | The kind of object in that scope. |
| Objects / Embeddings | How many objects there are, and how many have embeddings. |
| Searchable | Whether this scope is currently available to the AI. |
| Priority | Its importance relative to other scopes. |
| Last Status | The result of the last indexing run. |
| Last Indexed / Duration (ms) | When it was last refreshed and how long that took. |
| Updated | When the entry last changed. |
5. AI Runtime
The AI Runtime page shows the AI in action — the conversations it's handling and the suggestions it's putting forward.
5.1 RAG Sessions
A RAG session is one chat-style conversation with the AI (RAG is the technology behind the chat feature — it answers questions using your data). The RAG Sessions tab lists these conversations so you can review what was asked and answered.
Filter by Status: All, Active, Closed, or Expired.
| Column | What it shows |
|---|---|
| Session ID | The conversation's identifier. |
| Org / Domain | Who it belongs to. |
| Name | A label for the session. |
| Type | The kind of session. |
| User | Who was chatting. |
| Status | Active, Closed, or Expired. |
| Created | When it started. |
Row action: the Inspect icon (Open conversation) opens the full back-and-forth — every message in order, so you can see exactly what the AI was asked and how it replied. A message still being generated shows a Streaming… marker, and any that didn't complete show Interrupted or Error.
5.2 Suggestions
The Suggestions tab lists the ideas the AI has put forward — for example a suggested mapping — for a person to accept or reject. Each row is one suggestion, with how confident the AI is in it.
Filter by Status: All, Proposed, Accepted, Rejected, or Modified.
| Column | What it shows |
|---|---|
| Org / Domain | Who the suggestion is for. |
| Type | The kind of suggestion. |
| Source Type / Source Object | What the suggestion is about. |
| Suggestion | The AI's proposed answer. |
| Confidence | How sure the AI is, from 0 to 1 — higher is more confident. |
| Model | The model that produced it. |
| Status | Proposed, Accepted, Rejected, or Modified. |
| Created | When it was made. |
Row actions let you decide what happens to a proposal:
| Action | What it does |
|---|---|
| Accept | Approves the suggestion as-is. |
| Reject | Turns the suggestion down. |
| Modify | Opens the suggestion so you can adjust it before approving (see below). |
Modifying a suggestion opens a window showing the original proposal (its type, source, confidence, and model) and lets you enter:
- Modified Suggestion Text — your corrected version. This becomes a learning example the AI can reuse, so a good correction improves future suggestions.
- Modified Suggestion JSON (optional) — a structured version, if the suggestion needs one. It's checked to make sure it's valid.
- Reviewer Comment — a note saved alongside your change, explaining it.
YOUR CORRECTIONS TEACH THE AI — When you Modify a suggestion, your corrected text is kept as a canonical example and fed back in, so the AI learns from it. A clear, accurate correction is worth the extra moment — it makes the next suggestion better.
5.3 Suggestion Feedback
The Suggestion Feedback tab is the audit trail for the Suggestions tab — a record of every accept, reject, and modify, and who did it. You'd come here to see the history of decisions rather than to make them.
| Column | What it shows |
|---|---|
| Feedback ID | The identifier for this feedback entry. |
| Suggestion | The suggestion it relates to. |
| Org / Domain | Who it belonged to. |
| Action | What was done — accepted, rejected, or modified. |
| Comment | The reviewer's note, if any. |
| Modified Text | The corrected text, when the action was a modification. |
| User | Who took the action. |
| Created | When it happened. |
6. Quick Reference
A fast lookup for the most common actions.
| I want to… | Do this |
|---|---|
| Open AI Process | Process → AI Process → AI Knowledge or AI Runtime |
| See the AI's working contexts | AI Knowledge → AI Projects |
| Add a project | AI Projects → + Add Project → fill the form → Save |
| Retire a project without deleting it | AI Projects → Archive (and Reactivate later) |
| Confirm some data was taken in | AI Knowledge → Embedding Store → filter by Project |
| Check nothing is stuck | AI Knowledge → Embedding Queue → turn on Stuck only |
| Get a stuck item moving | Embedding Queue → Retry on the row |
| See what's searchable in a domain | AI Knowledge → Relevancy Index → filter by Data Domain |
| Read a past AI conversation | AI Runtime → RAG Sessions → Inspect (Open conversation) |
| Review the AI's proposals | AI Runtime → Suggestions |
| Approve or turn down a suggestion | Suggestions → Accept / Reject |
| Correct a suggestion and teach the AI | Suggestions → Modify → edit the text → save |
| See the history of suggestion decisions | AI Runtime → Suggestion Feedback |
Source: OnCoor Process Management product documentation, written for end users. For the latest screens and options, always refer to the in-app interface.