Engineer the change

SOCI Cockpit — System of Connected Intelligence — is where work is done. Each conversation is backed by a long-lived coding-agent process running on your machine. The chat-style transcript and the terminal pane are two views of the same session.

Cockpit brings together four things around that session: the context from a SOCK harness, the commands from your workspace's organ pack, the tools the agent can reach, and the controls your organisation has set.

The point is not a more elaborate one-off prompt. It is a repeatable engineering environment in which the selected system context and the workspace's method travel with the task, while the human operator can inspect both the readable transcript and the underlying terminal process.

Sessions and reattachment

  • One conversation maps to one long-lived terminal-backed agent process.
  • Returning to a session restores the conversation from the agent's own transcript.
  • The terminal view replays a bounded amount of scrollback; it is not a complete permanent log.
  • Starting a new conversation creates a new session rather than silently reusing an old process.

Grounding

A Cockpit session answers from a specific body of knowledge, and you choose which. The SOCK selector at the top of Cockpit says which published harness grounds this workspace's sessions, and the choice is explicit — MSOCK never guesses which knowledge produced an answer.

A session is bound to whatever was selected when it started, for its whole life. Changing the selection affects the next session, not the one already running.

See ground sessions in SOCK knowledge for the full behaviour.

Commands

Your workspace installs a versioned organ pack of lifecycle commands, all beginning with /ws-. They run the AIDLC lifecycle — ideation, decomposition, design, implementation, testing, review, shipping, certification, and learning — and they are the same commands SOCM Command Center's gates expect evidence from.

The commands cheatsheet inside Cockpit lists what is installed, grouped by lifecycle stage. A first-time hint points at the command surface, shows itself briefly, and then gets out of the way.

Intelligence inside explicit controls

SOCI separates the parts that are easy to blur together in an AI-assisted workflow:

Building blockRole in the workflow
CommandA named, reusable instruction for a recognised piece of lifecycle work—not an improvised prompt
Agent runtimeThe long-lived coding process that performs the work and exposes its terminal state
Organ-pack guidanceVersioned organisational instructions installed for the workspace
Product and workspace controlsThe selected SOCK context, available tools and connections, feature settings, and operator choices that bound the session

The model still reasons probabilistically. Repeatability comes from making the command, context, runtime, and controls explicit—not from treating model output as deterministic or automatically safe.

Models and thinking effort

ModelUse whenNotes
SonnetBalanced coding, explanation, and everyday debuggingThe default
OpusComplex architecture, difficult debugging, or broad refactoringOpt in for heavier reasoning
HaikuSpeed and narrow, low-risk tasks matter mostDoes not support thinking effort

Thinking effort has five levels — Low, Medium, High, X-High, Max — and defaults to High. You can change it before a session starts or while one is running. Use a higher level when deeper analysis is worth the extra time; use a lower one for narrow, reversible tasks.

Workspace artefacts

Cockpit can browse the artifacts/ folder in your workspace and preview what a session produced — documents, diagrams, and generated files — without leaving the application. These are files on your own machine, so they carry no comments or shared review state.

Integrations

Cockpit can connect to Jira and GitLab so lifecycle commands can read and write the work items and merge requests they refer to. Both connections belong to the signed-in member. See Jira and GitLab for identity, direction, and product boundaries.

The converged chat

A redesigned chat experience is available on a limited-availability basis and is off by default. An enterprise or workspace administrator enables Cockpit Converged Chat from the Admin Console Features screen for the workspaces that should have it.

Current boundaries

  • Cockpit runs on the desktop only.
  • Sessions are grounded by an explicit selection; there is no automatic search across every harness in the workspace.
  • Comments and shared review state belong to SOCM Command Center work items, not to workspace files.
  • A release or withdrawal of a SOCK harness deliberately takes priority over an in-flight assistant answer about that harness.