BETA · The AI workspace for coding

Chat with agents, edit code, and run autonomous missions in one place.

The power of the Claude Code CLI (Skills, MCP, tool use, file-based work) running continuously server-side and opening in any browser, phone included. Message the agent, step away, come back to everything still running. No SSH, no tmux.

Talk to us Explore products

arclight.vn

More Than Code home screen: choose MTC Chat, IDE, or TaskForce

Live session

Watch More Than Code work

You steer every turn: the agent proposes, you approve or reject each tool — and each choice leads somewhere different. This is a mock-up of the real UI; try the buttons.

Extract email validation helper

Jul 20, 11:11

Extract the email validation in waitlist.ts into a shared helper.

11:11

Thought

Pull isValidEmail into lib/validation.ts and reuse it — avoid duplicating the regex in contact.ts.

I’ll read waitlist.ts, add the helper, then update both call sites. Reading the file first:

Called tool:Readsrc/routes/waitlist.ts

{
  "file_path": "src/routes/waitlist.ts"
}

Action requires approvalEdit

Create  src/lib/validation.ts
Edit    src/routes/waitlist.ts  (import helper)

Approved. Writing the helper and updating both call sites.

Artifacts

2 files changed+14-6

src/lib/validation.ts

+ export const isValidEmail = (v: string) =>

+ /^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(v.trim());

src/routes/waitlist.ts

- if (!/^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(email))

+ if (!isValidEmail(email))

Done. The helper lives in lib/validation.ts; both waitlist.ts and contact.ts use it. Commit from the Git panel whenever.

Ask anything…

Ask firstClaude Opus 4.8

Built on the Claude Agent SDK
Self-hostable
Multi-user
SSO
PostgreSQL

From scattered CLIs to one place.

Claude Code CLI alone

More Than Code

Session storage

Claude Code CLI aloneLocal per machine

More Than CodeOne PostgreSQL, resume anywhere

Switch machines

Claude Code CLI aloneLose context, SSH in

More Than CodeOpen a browser, keep going

On mobile

Claude Code CLI aloneSSH drops, tmux won't swipe, session dies

More Than CodeInstall as an app (PWA), session stays alive, smooth

Stepping away

Claude Code CLI aloneClose the terminal, lose it

More Than CodeMessage, go do something else, come back running

Collaboration

Claude Code CLI aloneOne person, one terminal

More Than CodeMulti-user, isolated workspaces

Tool approval

Claude Code CLI aloneTerminal prompts

More Than CodeVisual approvals, rewind & edit turns

Editing code

Claude Code CLI aloneLeave for another editor

More Than CodeMonaco + ripgrep + git in-tab

Automation

Claude Code CLI aloneRoll your own scripts

More Than CodeHeadless Missions, checkpoints, auto-merge

Three ways to work. One platform.

MTC Chat

Pair with an agent, you stay in control.

Every session runs server-side and streams to your browser in real time. Approve each tool before it runs, answer the agent's questions, rewind or edit any turn.

Tool approvals: ask / safe / bypass modes.
Rewind & edit the conversation, review transcript versions.
Subagents shown first-class inside the transcript.
Git, Todo, Workflow, and Context panels on the right.

Learn more →

arclight.vn

Fix an N+1 query in the dashboard

Jul 22, 10:32

The dashboard loads slowly — check the loader for an N+1 query and fix it.

Thought

The loader likely runs a query per row — collapse into one query or preload by id.

Readsrc/dashboard/loader.tsdone

SubagentExploreFind loaders that query inside a loop3 activities

Globsrc/**/loader.ts

GrepfindBy|await .*repo

Readsrc/dashboard/repository.ts

Needs your approvalEdit

Edit  src/dashboard/loader.ts       (collapse N+1 → 1 query)
Edit  src/dashboard/repository.ts   (add findManyByIds)

RejectApprove for sessionApprove

Ask anything…

Ask firstClaude Opus 4.8

IDE

Edit your code right in the browser.

A Monaco editor, project-wide ripgrep search, diffs, git actions, and a real bash terminal, all without leaving the tab. The Chat tab embeds the same agent from MTC Chat.

Monaco: save, stale-file detection, Markdown/HTML/SVG/image preview.
Project-wide ripgrep search (Ctrl+Shift+F).
Diff viewer, git panel: stage/commit/push/pull, switch branch.
A real bash terminal over WebSocket.

Learn more →

arclight.vn

currency.ts

format.ts

const FORMATTERS = new Map<string, Intl.NumberFormat>();

/** Shared currency formatter for the checkout + invoice views. */
export function formatCurrency(amount: number, currency = 'USD'): string {
  let fmt = FORMATTERS.get(currency);
  if (!fmt) {
    fmt = new Intl.NumberFormat('en-US', { style: 'currency', currency });
    FORMATTERS.set(currency, fmt);
  }
  return fmt.format(amount);
}

export function formatCents(cents: number, currency = 'USD'): string {
  return formatCurrency(cents / 100, currency);
}

changes · currency.ts+6−2

−export const fmt = (n) => '$' + n.toFixed(2);

−export const fmtCents = (c) => fmt(c / 100);

+import { formatCurrency, formatCents } from '../lib/currency';

+return formatCents(order.totalCents, order.currency);

terminal

$ git status --short

M src/lib/currency.ts

M src/lib/format.ts▋

TaskForce

Hand off the work, the agent runs it.

A Mission is an ordered chain of Tasks that runs headless on its own git branch. Each file-changing Task creates a checkpoint commit to track and roll back step by step. Transient errors retry automatically; on finish it can auto-merge and let AI resolve conflicts.

Fresh context per Task; a DAG runs independent Missions in parallel.
Checkpoint commit per step; retries reset to the last checkpoint.
Structured JSON-schema output per Task.
Auto-merge with AI conflict resolution; alerts via Microsoft Teams / Web Push.

Learn more →

arclight.vn

TaskForceNew mission

Missions62 draft

Running12 queued

Completed11 failed

Total cost$4.0084k tok

Last run2m ago

Refactor billing module$0.422m ago3/5running

Nightly i18n sync$1.103h ago6 taskserrors

Migrate API to v2$2.305h ago8 taskscompleted

Fix flaky e2e tests$0.186h ago4 tasksfailed

#3Extract invoice calculator1m 12scompleted

key: billing.extract-invoice · model: claude-opus · 12,480 tok

3f9a1c2checkpointauto retry · reset to checkpointJSON SchemaAuto-merge: on

Structured output

{
  "filesChanged": 3,
  "testsPassed": true,
  "checkpoint": "3f9a1c2"
}

Dogfooding

More Than Code is built with More Than Code.

Every feature above is what our own team uses daily to build this very product: chatting with an agent to fix bugs, reviewing diffs and committing right inside the browser IDE, handing repetitive work to autonomous missions. Nothing ships here until we trust it ourselves.

Get going in three steps.

1
Connect a provider: Anthropic (OAuth) or your own API key + baseUrl.
2
Open your workspace: an isolated space per user.
3
Chat, edit code, or hand off an autonomous mission.

Your models, your extensions.

Model

Connect Anthropic via OAuth (subscription), or use an API key + custom baseUrl, including compatible proxies.

MCP

Add per-user MCP servers; a built-in git MCP handles fetch/pull/push.

Skills

Upload your Skills (.md/.zip/.skill), isolated per user.

Your data, your control.

Enterprise SSO (Azure AD, Google, GitHub)
Invite-based access, domain allow-list
Isolated workspace per user

See security details →

FAQ

What is More Than Code?

A browser-based AI workspace built on the Claude Agent SDK, with three products: MTC Chat (chat), IDE (editing), TaskForce (automation).

What does BETA mean?

The product is maturing and access is limited by invite. Talk to us if you want in early and help shape it.

How do the three products differ?

MTC Chat is live chat with the agent; IDE edits/searches code in the browser (embedding that same agent); TaskForce runs work unattended, headless.

Can enterprises self-host?

Yes, More Than Code is self-hostable. Talk to us about deployment.

Get started with More Than Code.

We're in BETA with invite-based access. Talk to us about what you need.

Talk to us