# Company Console guide

The Company Console is the browser workspace for operators. It reads and acts
on Company OS through the Company OS API; it does not keep its own copy of
company state. Use it to describe work, follow tasks, decide approvals, inspect
receipts, manage agents and connectors, and check system health.

> **Status.** The Console is the internal and operator web UI. The customer
> desktop shell is the [Company OS desktop app](./DESKTOP_APP.md). In
> `supported-capabilities.json` (as of 2026-09-16) the release gate is
> `blocked` and the strategy is an invite-only private preview. A route that
> exists in the Console is not thereby release-supported. See
> [Canonical product path](./CANONICAL_PRODUCT_PATH.md) and
> [Start here](./COMPANY_OS_START_HERE.md).

## Open the Console

From a source checkout, run one command from the repository root:

```bash
./company-os
```

It starts or reuses the local services, opens the Workspace start page, and
prints one URL. With default settings:

| Surface | URL |
| --- | --- |
| Company Console (Workspace) | `http://127.0.0.1:3050/workspace/start` |
| Company OS API | `http://127.0.0.1:3847` |
| API health | `http://127.0.0.1:3847/api/company/health` |
| AgentChat runtime health | `http://127.0.0.1:3050/api/agent-chat-runtime` |

Change the ports with `HSM_COMPANY_CONSOLE_PORT` (Console) and
`HSM_CONSOLE_PORT` (API).

| Command | Effect |
| --- | --- |
| `./company-os status` | Prints the Workspace URL when API, AgentChat, and Workspace are healthy. |
| `./company-os up --no-open` | Starts or reuses the stack without opening the browser. |
| `./company-os down` | Stops the services this checkout started. |

The launcher needs local Rust, Node/npm, reachable Postgres, and model access
for provider-backed turns. Full startup and troubleshooting steps are in
[Workspace and chat launch](./COMPANY_OS_AGENT_CHAT_LAUNCH.md). The routes `/`
and `/workspace` both redirect to `/workspace/start`.

## Sign in

When the Console cannot load the company list because the request is
unauthorized, it sends you to `/operator-login`, remembering the page you came
from.

1. Open the **Operator sign-in** page (`/operator-login`).
2. Choose **Continue on this device**. This signs you in without a key when the
   device is running Company OS locally.
3. If that is unavailable, choose **Use an operator key instead**, enter the
   **Operator key**, and select **Open Company OS**. The key is kept in an
   HttpOnly session cookie and is not exposed to workspace scripts.
4. After the operator session, the page establishes a human session for the
   device. If that fails it shows: "This device could not be signed in. Ask your
   administrator to authorize it."

After sign-in you return to the page you came from if it is under
`/workspace/` or is a CLI authorization page (`/cli/authorize`, `/cli/device`).
Otherwise you land on `/workspace/start`.

**Company access.** The company selector is in the page header. To join another
company, open **Join another company** and either enter a **Company ID or
address** (sends an access request) or an **Invitation link or code**. Owners and
admins see **Pending access requests** there.

## Layout

- **Sidebar.** Seven primary entries: Chat, Work, Beads, Proof, Agents,
  Connectors, Operate. Below them, **All workspace areas** expands the full
  directory of every other surface, grouped by job. The footer links to
  **System** (system health) and **Legacy console** (`/classic`).
- **Tab strips.** Work and Proof each show sibling views as tabs under the header.
- **Search.** Press `Cmd+K` (or `Ctrl+K`) or use the search button to search the
  company; the Search page lists what it covers.
- **Right rail.** A resizable rail that appears when you select a task or an
  agent and can be hidden from the header.
- **Overview.** The brand link at the top of the sidebar opens the dashboard.

Pages that need a company say "Select a company in the header" until one is selected.

## Main areas

### Chat (`/workspace/start`)

The front door. Tell the company what needs to happen; the conversation drives
the work and the result stays open beside it.

- The left pane is the chat. Starter prompts: "Plan this week's highest-impact
  work", "Review blocked tasks and propose next actions", "Research competitors
  and draft a launch memo". If you are unsure where to start, answer the three
  questions in the result pane and the page prepares one concrete workflow.
  Nothing runs until you choose **Use this workflow** or **Run first step**.
- The chat settings menu (the three-dots button) holds **Run controls**: create a
  company when none is active, **Create a new governed proof workspace**, a
  **Model and execution** choice (**Cloud** or **Local**), and the **Employee
  leading this work**.
- The result pane has three tabs: **Result**, **Evidence**, and **Task**.
  **Evidence** ("Run proof and verification") shows setup readiness (company
  API, model execution, worker execution, core MCP client, desktop bridge)
  separately from the outcome evidence for your request. **Task** shows the
  dispatched task, with **Open task** and **Review approvals** links.
- Header shortcuts open **Run receipts** and **Approvals**.
- If the run needs authority, the page offers **Connect owner or admin**,
  **Create governed workspace and retry**, or **Sign in as operator**.

### Work (`/workspace/issues`)

One task ledger. Tabs: **All tasks**, **Your tasks**, **Approvals**, **Beads**.

- **All tasks** lists tasks with filters **All**, **Active**, **Needs you**,
  **Blocked**, **Done**. **New task** opens a form. **Board** switches to a
  kanban view; **List view** switches back. A **Related work** panel appears when
  tasks share a project or workspace attachment; these links do not create
  dependencies.
- **Your tasks** (`/workspace/my-work`, page title "My Work") has four sections:
  **In Review**, **Assigned to Me**, **Recently Touched** (last 24 hours), and
  **Failed Runs**.
- **Approvals** (`/workspace/approvals`, page title "Inbox") is for human
  decisions only. It shows counts for total, tasks, email drafts, and failures,
  then **Tasks awaiting approval** (**Approve** or **Approve done**, **Reject**,
  **Open**), **Runs needing your attention**, **Email drafts to review**
  (**Approve draft**, **Reject**), and **Recent errors**. Approving an email
  draft does not send it; delivery is a separate, receipt-backed action. A
  **Human signing** panel covers passkey enrollment and approval.

### Beads (`/workspace/beads`)

The task graph and ready work, titled "Company OS Beads". It shows ready ledger
rows, triage lists (top picks, quick wins, blockers to clear), graph and
analytics views, and a **Robot JSON** export link. It is also the fourth Work tab.

### Proof (`/workspace/quality`)

Evidence about what happened. Tabs: **Quality**, **Diagnostics**,
**Calibration**, **Lessons**, **Run receipts**, **Trace explorer**, **Papercuts**.

- **Quality**: select agents, run quality checks, and use trace-backed evidence
  to decide what can be promoted (quality packs, agent KPIs, run stories,
  promotion gates).
- **Diagnostics** (`/workspace/diagnostics`): the diagnostic inbox of analyzer
  findings with proof. Each finding links to its source revision, run receipt,
  responsible task, and raw analyzer output when present. Lifecycle state
  changes only when a producer submits a complete snapshot.
- **Calibration** (`/workspace/quality/decision-calibration`): decision
  calibration.
- **Lessons** (`/workspace/learn`, page title "Recurring overrides"): decisions
  that people repeatedly changed, listed for review.
- **Run receipts** (`/workspace/runs`, page title "Verified runs"): recent agent
  runs with status, missing evidence, final summary, and approval state. Opening
  a run (`/workspace/companies/<company>/runs/<run>`) shows live attach, run
  surfaces, and the verified timeline: node state, evidence, spend, approvals,
  snapshots, rewind points, forks, and legal actions.
- **Trace explorer** (`/workspace/trace`): live agent runs as span trees, and
  fleet-level failure patterns.
- **Papercuts** (`/workspace/papercuts`): the inbox of recurring friction.
  Filter aggregates, inspect provenance, then resolve or promote open evidence
  to a task.

### Agents (`/workspace/agents`)

The team roster: who is active, what they know, and where to open their
workspace. Filter by **All**, **Active**, **Paused**. An agent page
(`/workspace/agents/<id>`) has tabs **Workspace**, **Memory**, **Skills**,
**Overview**, **Chat**, **Runs**, and **Computer**.

### Connectors (`/workspace/credentials`)

API secrets, linked accounts, triggers, and discovery in one place. Tabs:
**API secrets**, **Accounts & automations**, **Executor catalog**,
**Discovery**, **Plugin config**. Provider keys and connection tests are also on
**Models** (`/workspace/models`). For desktop account sign-in see
[Desktop connector sign-in](./DESKTOP_CONNECTOR_SIGN_IN.md).

### Operate (`/workspace/operate`)

The control-plane directory, titled "Operate the company". It links to
Governance, System health, and Proof, then lists every other surface in seven
groups (the same groups appear under **All workspace areas**). Review and proof
holds only Lessons (`/workspace/learn`).

**Work views**

| Label | Route | Purpose |
| --- | --- | --- |
| Overview | `/workspace/dashboard` | Work graph by status and priority, **Needs attention**, API health, shortcuts. |
| Recovery | `/workspace/recovery` | Track approved retry and replanning actions; proposals are reviewed through task approval first. |
| Outcome Radar | `/workspace/radar` | Ranked, evidence-backed proposed next moves. Proposal-only, no external action. |
| Scenario rehearsal | `/workspace/scenarios` | Forecasts and decision rehearsal (page title "Scenario Lab"). |
| Voice intake | `/workspace/voice` | Capture spoken intent, review the cleaned command, choose one action, persist the evidence. Voice is input, not authority. |
| Diff review | `/workspace/review` | Inspect publication evidence and add annotations. Approval stays in Approvals. |
| Beads | `/workspace/beads` | See above. |
| Search | `/workspace/search` | Search tasks, agents, projects, goals, memory, run activity, connectors, routes, and recent searches. |
| Files | `/workspace/files` | Inspect files, parse documents, and prepare evidence before an agent uses data. Upload is supported. |
| Website studio | `/workspace/web-clone` | Source-first website workflow. |
| Presentation studio | `/workspace/openvid` | Record, edit, and import presentation video. |

**Automation**

| Label | Route | Purpose |
| --- | --- | --- |
| Daemons | `/workspace/daemons` | Versioned `DAEMON.md` automation packages. |
| Daemon flows | `/workspace/workflows` | Multi-step work (page title "Workflows"): Explore gathers facts, Plan defines the approach, completed plans move to Build. Detail pages: `/workspace/workflows/<id>`. |
| Daemon runs | `/workspace/routines` | Runs, env, and coalescing. |
| One-click tasks | `/workspace/automations` | Reports, triage, quality scans. |
| Admission | `/workspace/mission-control` | "What can start now?": readiness, capacity, budgets, runs, and receipts. |
| SOPs | `/workspace/playbooks` | Author procedures and implement them as tasks. |
| Context canvas | `/workspace/context-canvas` | Branch from documents, receipts, and tasks. |

**Access and governance**

| Label | Route | Purpose |
| --- | --- | --- |
| Governance | `/workspace/governance` | Administrative authority, delegation expiry warnings, enforcement policy, principal tokens (shown once), and delegations. |
| System prompts | `/workspace/prompts` | Versioned prompt layers, review, and activation gates. |
| Grants | `/workspace/permissions` | Protected assignment permissions. |
| Connectors | `/workspace/credentials` | See above. |
| Financial agents | `/financial-agents` | Directory of financial APIs, MCP gateways, SDKs, and market data. |
| Extensions | `/workspace/extensions` | Stage signed packages and manage capability grants. |
| Models | `/workspace/models` | Provider keys (encrypted when `HSM_COMPANY_CREDENTIALS_KEY` is set) and connection tests. |

**Team**: Together (`/workspace/together`, invite a partner, follow the work, and review
together), Agents, Coordination (`/workspace/coordination`, durable agent
channels), Hire (`/workspace/hire`), Squads (`/workspace/squads`), Council
(`/workspace/council`), and Skills (`/workspace/marketplace`, install company
packs and skills).

**Data and memory**: Business Data (`/workspace/business-data`), Sync
(`/workspace/sync`), Map (`/workspace/graph`), Activity
(`/workspace/intelligence`), News digest (`/workspace/news`), and Dialogue
branches (`/workspace/dialogue`).

**Spend and system**

| Label | Route | Purpose |
| --- | --- | --- |
| Spend | `/workspace/costs` | Recorded month spend, by category, raw data; compare with agent budgets on Agents. |
| System health | `/workspace/runtime` | Local services, queues, recent errors, sandbox and mesh panels. |
| Notify | `/workspace/notifications` | Slack and Discord notifications. |
| Settings | `/workspace/settings` | Execution lane (keep on **Auto**), **Lab Mode**, code font preferences. |

The remaining entries in this group are lab surfaces; see the next section.

## Experimental and lab routes

These routes are not part of the supported path. No page marks most of them
"experimental"; the basis is that they are outside the main navigation and
outside the [canonical product path](./CANONICAL_PRODUCT_PATH.md) and
`supported-capabilities.json`. Treat them as development or lab surfaces.

**Listed only in the Operate directory (no primary navigation):**

- `/workspace/pixels` Agent Pixels: pixel operations map for agents and tasks.
- `/workspace/openvid` Presentation studio, and `/workspace/web-clone` Website studio.
- `/workspace/rust-space` Build and disk: Rust, Cargo, and workspace build storage.
- `/workspace/dead-star` Dead Star: preserved contributors and a local royalty
  ledger; settlement happens outside Company OS.
- `/workspace/browser-lab` Browser Agent Lab. This page labels itself "Lab only:
  no connector secrets, no company writes, no trust promotion."
- `/workspace/fleet` Fleet: opted-in installs reporting health telemetry.
- `/workspace/architecture` (Blueprint) and `/workspace/~architecture`
  (Architecture map).

**Shown only with Lab Mode on** (Settings, sidebar entries gated by feature
flags; some need a connected runner or provider):
`/workspace/snapshots`, `/workspace/promotion`, `/workspace/gepa` (needs optimizer
configuration), `/workspace/dynamic-plans`. Lab output is candidate evidence until
a verifier rerun and explicit human approval promote it.

**Routable but not linked from the navigation:** `/workspace/quality/eval-pack`,
`/workspace/quality/eval-runs`, `/workspace/quality/reflex`,
`/workspace/quality/review`, `/eval-setup`, `/workspace/companies/new` (Company
Passport onboarding, linked from the legacy console), and
`/investor-demo/accept` and `/investor-demo/workspace` (hosted sandbox for
investors; see [Investor demo mode](./INVESTOR_DEMO_MODE.md)).

**Legacy:** `/classic`, labelled **Legacy console** in the sidebar footer.

## CLI authorization pages

`/cli/authorize` and `/cli/device` are the browser half of `companyos login` and
`companyos login --device`. They are not part of the sidebar.

1. Run the login command in your terminal.
2. In the browser, sign in on this device if asked. For `/cli/device`, enter the
   code shown in the terminal (`ABCD-EFGH` form).
3. Choose the company. Only an owner or admin on a device authorized for that
   company can grant access; otherwise the page offers **Request access**.
4. Review the permissions and select **Authorize CLI with this device**. The
   fixed permission set is: Tasks, Memory, Customers, Connectors, Tools,
   Workspace files, Validation.

The grant is a separate CLI session for that company. Return to the terminal; it
finishes signing in. End it with `companyos logout`. If the page says the link
cannot be used, run the login command again for a fresh link. Commands and
options are in the [CLI reference](./CLI_REFERENCE.md); connecting agents is in
[Connect agents](./CONNECT_AGENTS.md).