Watch your agents work

Build plan for agent presence, cursors, suggestions and comments in doc and sheet artifacts

A

What ships

:::callout tip Agents show up in doc and sheet artifacts the way the affectUI demo shows. They sit in the presence stack and have labelled cursors with a task line and cost. Their edits arrive as suggestions you accept or reject, with Accept all, and their questions post under their own name. It ships on Desktop, the ai.allternit.com phone layout and the m.allternit.com PWA, with docs. :::

B

Where each piece goes

Piece	Repo	Files
Suggestions table, agent presence and comment fields (migration 096)	allternit	`cmd/allternit-cloud-api/migrations_pg/096_artifact_agent_watch.sql`
Suggestions routes, agent presence, agent comments	allternit	`routes/artifacts_suggestions.rs`, `routes/artifacts_comments.rs`
`artifact_suggest` tool, agent heartbeat with task and cost	allternit	`gizzi-code/.../builtins/artifact.ts`, `integrations/artifacts/index.ts`
Presence stack with agents, cursor layer, suggestion bar and popover, live refresh	allternit-ai	`components/artifacts/collab/*`, `ArtifactWindow.tsx`, `lib/artifacts/editors/suggestions.ts`
Tracked changes under the agent name in Cowork Docs	allternit-ai	`office-docs-app` AiPanel / pane executor
Guide, tool and API docs, dependency map	allternit	`surfaces/docs/...`, `docs/dependency-map`

C

How it runs

D

Order

Three workers write code in their own worktrees: cloud-api, gizzi, and the UI. Then docs.
I build once and run the checks. Then a smoke boot of cloud-api and one merge per repo.
Before prod migration 096, the PWA deploy and the Desktop runtime publish, I ask for your go.

---
template: sheet
title: Watch your agents work
subtitle: Build plan for agent presence, cursors, suggestions and comments in doc and sheet artifacts
---

## What ships

:::callout tip
Agents show up in doc and sheet artifacts the way the affectUI demo shows. They sit in the presence stack and have labelled cursors with a task line and cost. Their edits arrive as suggestions you accept or reject, with **Accept all**, and their questions post under their own name. It ships on Desktop, the ai.allternit.com phone layout and the m.allternit.com PWA, with docs.
:::

## Where each piece goes

| Piece | Repo | Files |
|---|---|---|
| Suggestions table, agent presence and comment fields (migration 096) | allternit | `cmd/allternit-cloud-api/migrations_pg/096_artifact_agent_watch.sql` |
| Suggestions routes, agent presence, agent comments | allternit | `routes/artifacts_suggestions.rs`, `routes/artifacts_comments.rs` |
| `artifact_suggest` tool, agent heartbeat with task and cost | allternit | `gizzi-code/.../builtins/artifact.ts`, `integrations/artifacts/index.ts` |
| Presence stack with agents, cursor layer, suggestion bar and popover, live refresh | allternit-ai | `components/artifacts/collab/*`, `ArtifactWindow.tsx`, `lib/artifacts/editors/suggestions.ts` |
| Tracked changes under the agent name in Cowork Docs | allternit-ai | `office-docs-app` AiPanel / pane executor |
| Guide, tool and API docs, dependency map | allternit | `surfaces/docs/...`, `docs/dependency-map` |

## How it runs

```flow
Agent edit -> artifact_suggest -> cloud-api stores pending suggestion
cloud-api stores pending suggestion -> Artifact window polls (4 s while agents are here)
Artifact window polls (4 s while agents are here) -> You Accept or Reject
You Accept or Reject -> Client applies change and saves one version
```

## Order

1. Three workers write code in their own worktrees: cloud-api, gizzi, and the UI. Then docs.
2. I build once and run the checks. Then a smoke boot of cloud-api and one merge per repo.
3. Before prod migration 096, the PWA deploy and the Desktop runtime publish, I ask for your go.