The integrated IG-XL agent
See what the IG-XL agent will read before you ask, and change it. Type your question and the curator ranks your project against it: program sheets and VBA, requirements, production lots, load boards, the channel plan. Each item shows why it matched. Switch items on or off, pin the ones that always matter, and hand the selection to the agent in one click.
Type the task and score it. Every candidate appears with its kind, its score and the words it matched on. When a file in your own code is picked, the local files it depends on come with it, marked as pulled in by dependency. Switch off what does not belong and hand off. That selection is the project context the agent receives, and the chat lets you open it before you send.
Scoring is deterministic: same project, same question, same ranking. It makes no model call, so you can re-score as often as you like while you refine the question, at no model cost. The model is called only in the chat, with the context you approved.
| What it ranks | What reaches the agent |
|---|---|
| IG-XL sheets and VBA modules | The program's sheets and procedures, from a synced repository or from the workbook the agent reads in Excel. |
| Code | Files in synced repositories, with the local files they depend on. |
| Requirements | The project's spec requirements. |
| Production lots | The project's STDF lots. |
| Load boards | The project's board designs. |
| Channel plan | The saved channel plan. |
| Generated outputs | What ATE·IQ has generated for the project. |
Every scored item carries a toggle. Flip one and the assembled preview updates with it, so you watch the context change as you change it. Per-kind bulk toggles flip a whole class in one click: all requirements off, all generated outputs on.
Pins are for the items you always want. A pinned item survives re-scores: even when a new question's scoring drops it from the visible list, it still travels with your selection. Pins persist per project across reloads. One rule keeps intent unambiguous: an item you explicitly switched off stays off, pin or no pin.
Two ceilings keep the project context in check, and both render as live meters in the preview. When a cut is needed it is disclosed: you are told what went, not left to guess.
| Ceiling | Size | At the limit |
|---|---|---|
| Auto-assembly budget | 48,000 characters | Applies when the curator assembles from its own scores. The lowest-ranked sections (task intent first, then keyword score) shrink or drop first; dropped sections are named in the preview and trimmed items carry a budget-trimmed badge. |
| Handoff cap | 64 KB | Applies to a selection you hand off. An over-cap handoff is refused at the button, showing the assembled size against the cap, and the chat enforces the same ceiling. |
The handoff is one click. The selection is saved as a draft the chat picks up on its own: it survives a reload and a new tab, prefills the composer with your task, expires after 24 hours, and is used up only by the first send that succeeds. A send that fails leaves the draft in place.
The curator is how the IG-XL agent gets its project context wherever the question needs the whole project, and it steps aside where the open workbook is enough.
| Where you ask | How the context is chosen |
|---|---|
| Desktop chat | The curator assembles project context for each question, or the chat uses the selection you handed off. |
| Agent pane beside Excel, Debug mode | The curator adds project context, and the open workbook's live sheets and VBA are used in place of the project's stored copies. |
| Agent pane beside Excel, Edit mode | The open workbook is the context. The curator is skipped to keep edits fast. |
| Code Map | Ask about a file and the curator assembles context with that file preselected, then hands it to the chat. |