Coworker IA: agents are principals, DMs are 1:1, channels are rooms

DESIGN.md

@@ -16,10 +16,11 @@ Top to bottom:
 
 1. **Brand row** — logo mark and a "New workbench" button (`+`) that
    opens `/new`.
-2. **Bench list** — the "Workbenches" label, then rows of workbench
-   conversations, with search built into the list itself. Nothing
-   page-scoped ever renders in this body; it lists conversations, not
-   product sections.
+2. **Bench list** — **Agents**, then **Channels**. Agent rows are
+   `kind: chat` DMs (one per agent). Channel rows are `kind: workbench`
+   rooms. Search is built into the list itself. Nothing page-scoped
+   ever renders in this body; it lists conversations, not product
+   sections.
 3. **Footer rail** — Mission Control is pinned above the rail as its own
    row. The first-run rail below it is Routines, Files, Skills, Agents, in
    that order (CL-6765). Insights and Evals join that rail only when
@@ -35,13 +36,14 @@ Top to bottom:
    direct one-click control to Settings, not a menu item — Settings
    never cost two clicks to reach.
 
-A workbench is an agent conversation, and the bench list IS the switcher —
-its rows are the primary way to move between workbenches, with no separate
-"switcher" control layered on top. The command palette's hidden "Switch
-workbench" action is a second door onto the same list, reachable by search
-rather than by scanning rows; it does not replace the sidebar as the
-switching mechanism. Approvals render inside the conversation, never as a
-standing band in the shell.
+A workbench is a conversation tenant — a DM with one agent, or a
+channel with many people and agents. The bench list IS the switcher —
+its rows are the primary way to move between those conversations, with
+no separate "switcher" control layered on top. The command palette's
+hidden "Switch workbench" action is a second door onto the same list,
+reachable by search rather than by scanning rows; it does not replace
+the sidebar as the switching mechanism. Approvals render inside the
+conversation, never as a standing band in the shell.
 
 ## Pages & Routing
 
@@ -59,8 +61,11 @@ behind at its old path — old links and bookmarks always land somewhere
 real, never a 404.
 
 **`/new` is the shipped create surface** (`NewWorkbenchPickerRoute`). The
-primary act is a prompt: say what the workbench should do, or pick a
-template shortcut underneath. The sidebar `+` and an empty-bench land on
+primary act is a prompt: say what the channel should do, or pick a
+named-template shortcut underneath. Blank `+` / prompt mint an empty
+channel and invite nobody. Named templates mint that same empty
+channel, then invite existing agents (including Myra as a participant,
+never as the mint host). The sidebar `+` and an empty-bench land on
 `/` both open this route — there is no Describe door.
 
 **`/inbox` is gone as a page** (CL-6151). The path stays as a redirect

PRODUCT.md

@@ -7,16 +7,18 @@ Platform, built on [Interchange](https://github.com/faremeter/interchange).
 
 ## The one concept
 
-Everything in Workbench collapses to a single idea: **a workbench is an
-agent conversation.** Opening a workbench means opening a conversation —
-with one agent, or with a group of people and agents — and that
-conversation is also its own tenant, so its membership and grants are its
-own. There is no separate "project" or "space" object sitting above the
-conversation; the conversation is the unit of work.
-
-This is why the sidebar is one list of workbenches, not sections split by
-kind — every workbench a person has, agent conversation or group
-conversation alike, shows up the same way. See
+Everything in Workbench collapses to a single idea: **a workbench is a
+conversation tenant.** People and agents are both principals
+(Interchange). Opening an agent opens the one 1:1 DM with that agent
+(`kind: chat`) — two clicks never clone a second DM. Opening a channel
+opens a multi-principal room (`kind: workbench`). There is no separate
+"project" or "space" object sitting above the conversation; the
+conversation is the unit of work.
+
+This is why the sidebar is Agents then Channels, not one recency list
+titled Workbenches. The same Interchange agent can sit in its DM and in
+many channels; product reopens or invites, and does not clone the
+definition or mint a sibling instance per room. See
 [docs/GLOSSARY.md](docs/GLOSSARY.md) for the full term mapping and
 [docs/CHAT.md](docs/CHAT.md) for how a conversation is built underneath.
 
@@ -32,19 +34,18 @@ team spaces); within a bench, they open, create, and work in workbenches.
 Workbench is intentionally not a multi-pane IDE. The product surface is one
 column at a time:
 
-- **A sidebar of workbenches** lists every conversation in the selected
-  bench, flat, most-recently-active first. There is no separate "channels"
-  vs. "chats" grouping the sidebar exposes to a person — every row is a
-  workbench.
-- **"+ New Workbench" always creates.** It opens `/new` — the shipped
+- **A sidebar of Agents then Channels** lists conversations in the
+  selected bench. Agent rows are DMs (`kind: chat`); channel rows are
+  rooms (`kind: workbench`). There is no one recency list titled
+  Workbenches.
+- **"New workbench" always creates.** It opens `/new` — the shipped
   prompt-primary picker — never a picker of existing things to join.
-  Starting a new workbench is the one way in, whether the result is a
-  one-on-one conversation with an agent or a group conversation
-  with people and agents together.
-- **Agents are templates.** Starting a new agent conversation means picking
-  an agent definition (a named, reusable capability) as the starting point
-  — the same definition can be launched into any number of separate
-  conversations, each with its own history and its own tenant.
+  Plus mints an empty channel. Nobody is auto-hosted.
+- **Agents are principals, not templates.** Opening Sales opens Sales —
+  the one 1:1 tenant with that agent. The same agent can sit in its DM
+  and in many channels. Product reopens or invites; it does not clone
+  the definition or mint a sibling instance per room. Myra is an agent
+  row, not a special home slot.
 - The active workbench occupies the main column; a contextual panel beside
   it carries account-wide surfaces (approvals, recent activity) that stay
   visible regardless of which workbench is open.
@@ -57,17 +58,19 @@ lands on the create surface rather than an empty shell:
 1. **Login.**
 2. **Credential** — connect a model provider (one-click OAuth for
    supported providers, or a pasted API key); see `packages/onboarding`.
-3. **Create** — `/` hops an empty bench to `/new`, the prompt-primary
-   picker (`apps/web/src/pages/new-workbench-picker.tsx`). A prompt box is
-   the primary act: typing a goal and submitting creates a blank
-   workbench and sends that text as the first message. Prefab template
-   rows underneath are one-click shortcuts, not a kind-then-Create
-   second step. There is no Describe door and no
+3. **Create** — `/` hops an empty bench (zero workbenches) to `/new`, the
+   prompt-primary picker (`apps/web/src/pages/new-workbench-picker.tsx`).
+   A prompt box is the primary act: typing a goal and submitting mints an
+   empty channel and sends that text as the first message; blank plus
+   invites nobody. Named-template rows underneath mint that same empty
+   channel, then invite existing principals (including Myra as a
+   participant, never as mint `definitionId`) — one-click shortcuts, not
+   a kind-then-Create second step. There is no Describe door and no
    `describe-first-workbench.tsx`.
 
 A bench that already has one or more workbenches skips create and lands
-in an existing conversation instead (see
-`apps/web/src/pages/home-page.tsx`).
+on `workbenches[0]` (see `apps/web/src/pages/home-page.tsx`). Myra is an
+agent row, not a home slot.
 
 The shell's first-run destinations stay small on purpose (CL-6765):
 Mission Control is pinned above the footer rail; the rail itself is
@@ -154,23 +157,24 @@ as an explicit absence, never a fabricated zero. See `packages/insights`.
 
 User-facing surfaces (UI, docs, support) use exactly these nouns:
 
-- **Workbench** — a single conversation, one-on-one with an agent or a
-  group of people and agents. Never "channel" or "bench" in user-facing
-  copy.
-- **Space** — the user-facing name for a group conversation surface (a
-  workbench with more than one counterpart).
-- **Chat** — the user-facing name for a one-on-one conversation (with an
-  agent or with a person).
+- **Workbench** — the product name, and the mint verb ("New workbench").
+  A workbench is a conversation tenant: a DM or a channel.
+- **Agent** — a coworker identity in the sidebar's Agents list. Opening
+  the row reopens that agent's one DM. Never "template."
+- **DM** — the one 1:1 conversation with an agent. Never cloned by a
+  second open.
+- **Channel** — a multi-party room. Plus mints an empty one; nobody is
+  auto-hosted. Named templates invite existing agents into that room.
 - **Bench** — the shared team scope a person signs into and switches
   between; shown in the bench switcher, never called a "workspace" or
   "org" in copy.
 
-Internally these map onto platform primitives (tenant, DM) — see
-[docs/GLOSSARY.md](docs/GLOSSARY.md) for the authoritative table. Code and
-API paths generally keep the platform's own names; "workbench" is the one
-exception (CL-6260), since its package (`@corbits/chat`) is ours, not the
-platform's — only user-facing surfaces use the rest of the product
-vocabulary above.
+Internally these map onto platform primitives (principal, tenant, kind:
+chat vs kind: workbench) — see [docs/GLOSSARY.md](docs/GLOSSARY.md) for
+the authoritative table. Code and API paths generally keep the
+platform's own names; "workbench" is the one exception (CL-6260), since
+its package (`@corbits/chat`) is ours, not the platform's — only
+user-facing surfaces use the rest of the product vocabulary above.
 
 ## Open questions
 
@@ -179,10 +183,6 @@ vocabulary above.
   succeeded) stays **target** until CL-6737 and CL-6738 land — see
   IMPLEMENTATION.md open questions; do not document those guarantees as
   shipped.
-- Whether "Space" and "Chat" as user-facing labels are fully rolled out
-  across the UI or still landing incrementally is not settled in the
-  docs reviewed for this pass — treat the sidebar and conversation-header
-  copy as the source of truth over this document if they disagree.
 - The precise boundary of what Insights surfaces to a non-admin bench
   member (all tenant activity vs. only their own) is not spelled out in
   `packages/insights`'s own docs as of this writing.

apps/web/src/app.css

@@ -3105,6 +3105,13 @@ tr.insights-row-clickable:hover {
   color: var(--muted-foreground);
 }
 
+.shell-panel-list-empty {
+  margin: 0 0 0.35rem;
+  padding: 0.15rem 0.4rem 0.35rem;
+  font-size: 0.78rem;
+  color: var(--muted-foreground);
+}
+
 /* CL-6224: per-workbench Timeline — one wall-clock spine (chat messages,
    thread forks, routine runs, tasks, approvals), day dividers, and a right
    rail of per-day KPI counts. Matches the Insights stage's own visual

apps/web/src/command-palette-actions.test.ts

@@ -35,7 +35,8 @@ function context(overrides: {
   const ctx = {
     path: overrides.path,
     navigate: (to: string) => navigated.push(to),
-    tenantId: overrides.tenantId ?? "tenant-1",
+    tenantId:
+      overrides.tenantId !== undefined ? overrides.tenantId : "tenant-1",
     cycleTheme: () => {
       themeCycled = true;
     },
@@ -73,8 +74,8 @@ describe("ACTION_COMMANDS", () => {
   test("new-workbench and new-agent speak consumer language, not mint", () => {
     const workbench = ACTION_COMMANDS.find((c) => c.id === "new-workbench");
     const agent = ACTION_COMMANDS.find((c) => c.id === "new-agent");
-    expect(workbench?.subtitle).toBe("Start a new workbench with Myra");
-    expect(agent?.subtitle).toBe("Start a new workbench with Myra");
+    expect(workbench?.subtitle).toBe("Start a new workbench");
+    expect(agent?.subtitle).toBe("Start a new workbench");
     expect(workbench?.subtitle.toLowerCase()).not.toContain("mint");
     expect(agent?.subtitle.toLowerCase()).not.toContain("mint");
   });
@@ -156,6 +157,65 @@ describe("runActionCommand", () => {
     await runActionCommand("talk-to-myra", ctx);
     expect(navigated).toEqual([]);
   });
+
+  test("talk-to-myra opens Myra's DM via kind=chat + definitionId, not a title-match mint", async () => {
+    const calls: { readonly path: string; readonly init?: RequestInit }[] = [];
+    globalThis.fetch = ((input: RequestInfo | URL, init?: RequestInit) => {
+      const path =
+        typeof input === "string" ? input : new URL(String(input)).pathname;
+      calls.push(init === undefined ? { path } : { path, init });
+      if (path.includes("/workflows/definitions")) {
+        return Promise.resolve(
+          new Response(
+            JSON.stringify({
+              data: [
+                {
+                  id: "wfd_assistant",
+                  tenantId: "tenant-1",
+                  name: "assistant",
+                  currentVersion: "1",
+                  status: "deployed",
+                  createdAt: "2026-01-01T00:00:00.000Z",
+                  updatedAt: "2026-01-01T00:00:00.000Z",
+                },
+              ],
+              nextCursor: null,
+            }),
+            { status: 200, headers: { "content-type": "application/json" } },
+          ),
+        );
+      }
+      if (path.endsWith("/chat/workbenches")) {
+        return Promise.resolve(
+          new Response(
+            JSON.stringify({
+              id: "chan_myra_dm",
+              title: "Myra",
+              kind: "chat",
+              pinned: false,
+              participants: [],
+            }),
+            { status: 200, headers: { "content-type": "application/json" } },
+          ),
+        );
+      }
+      throw new Error(`unexpected fetch: ${init?.method ?? "GET"} ${path}`);
+    }) as typeof fetch;
+
+    const { ctx, navigated } = context({ path: "/" });
+    await runActionCommand("talk-to-myra", ctx);
+
+    const createCall = calls.find((call) =>
+      call.path.endsWith("/chat/workbenches"),
+    );
+    expect(createCall?.init?.method).toBe("POST");
+    expect(JSON.parse(String(createCall?.init?.body))).toEqual({
+      kind: "chat",
+      definitionId: "wfd_assistant",
+      reuseExisting: true,
+    });
+    expect(navigated).toEqual(["/w/chan_myra_dm"]);
+  });
 });
 
 // Backs the `>` command palette's "New routine" and "Make this a routine"

apps/web/src/command-palette-actions.ts

@@ -28,9 +28,11 @@
 // never hops to `/routines` first the way this used to.
 
 import { createPendingDialogRequest } from "@corbits/shell-layout";
-import { WORKBENCH_PATH_PREFIX, workbenchPath } from "./workbench-path";
+import { WORKBENCH_PATH_PREFIX } from "./workbench-path";
 import { NEW_WORKBENCH_PATH } from "./routes";
-import { ensureMyraWorkbench } from "./myra-workbench";
+import { listAgentDefinitions } from "./agents-api";
+import { openAgentDmChat } from "./agent-dm-launch";
+import { findMyraDefinition } from "./myra-workbench";
 import { requestLibraryUpload } from "./library-upload";
 import type { RoutinePanelSubject } from "./shell/canvas-availability";
 
@@ -96,12 +98,12 @@ export const ACTION_COMMANDS: readonly ActionCommand[] = [
   {
     id: "new-workbench",
     title: "New workbench",
-    subtitle: "Start a new workbench with Myra",
+    subtitle: "Start a new workbench",
   },
   {
     id: "new-agent",
     title: "New workbench",
-    subtitle: "Start a new workbench with Myra",
+    subtitle: "Start a new workbench",
   },
   {
     id: "new-routine",
@@ -194,9 +196,9 @@ export async function runActionCommand(
     }
     case "talk-to-myra": {
       if (ctx.tenantId === null) return;
-      const result = await ensureMyraWorkbench(ctx.tenantId);
-      if (result.kind === "ready")
-        ctx.navigate(workbenchPath(result.workbenchId));
+      const myra = findMyraDefinition(await listAgentDefinitions(ctx.tenantId));
+      if (myra === undefined) return;
+      await openAgentDmChat(ctx.tenantId, myra.id, ctx.navigate);
       return;
     }
     case "go-workbenches": {

apps/web/src/command-palette-provider.tsx

@@ -451,9 +451,10 @@ export function CommandPaletteProvider({
     [artifactsQuery],
   );
 
-  // Order matches the mock's buildCmdkEntries: Commands, Workbenches,
-  // Pages, then the unscoped catalogs (Runs, Routines, Skills, Library),
-  // with People & agents last among the palette's groups.
+  // Order matches the mock's buildCmdkEntries: Commands, Agents &
+  // Channels, Pages, then the unscoped catalogs (Runs, Routines,
+  // Skills, Library), with People & agents last among the palette's
+  // groups.
   const sources = useMemo<readonly PaletteSource[]>(
     () => [
       {
@@ -464,7 +465,7 @@ export function CommandPaletteProvider({
       },
       {
         id: "workbenches",
-        heading: "Workbenches",
+        heading: "Agents & Channels",
         kind: "workbenches",
         items: workbenchItems,
       },