Data Curation Environment (DCE)
Cycle-based AI development, in the editor you already live in.
The DCE turns AI-assisted work into cycles: you describe what this round should accomplish, activate your agent, and review what came back — then write the next cycle. Your context is curated, your project's memory is artifact-first, and the whole loop is visible in one place: the Ascentia world.
Built for people who use VS Code + Claude Code with a Claude subscription. Your own agent does the work; the DCE curates what it sees and keeps the history.

The loop
- Select the context. Check the files that matter in the File Tree View — code, docs, design artifacts. The DCE flattens your selection (plus your cycle history) into one
agent.md the agent reads top-down.
- Write the cycle. Say what this round should do, in plain words. Cycle Notes color your draft as you write — green where a thought is named, red where it isn't yet.
- Activate the agent. One click hands the cycle to Claude Code (or your harness of choice). No copy-paste, no prompt assembly.
- Watch the work. The Ascentia world visualizes the live session — transmissions, threads, token flow, the orb streaming. Your past cycles orbit as a constellation.
- Review by git colors. The agent's work lands uncommitted. Modified and new files in your tree ARE the review; the pinned response carries the summary, the diffs, and the handoff list. Keep it and write the next cycle — or
git restore . and try again.
What you get
- Context curation — file selection with token counts, shadow-repo support, binary/PDF/Office extraction, and automatic selection upkeep as the agent proposes additions.
- Cycle history as memory — every cycle's context, response, and changed files persist locally; sprints group into orbits you can revisit.
- The Ascentia world — a live visualization of your agent at work: transmission log, thread telemetry, approval and question mirrors, the Code Cosmos of your repo.
- Cycle Notes — real-time key-pair structure over your draft, with AI-suggested keys grounded in your own workspace.
- Dictation — streaming speech-to-text into the cycle context. No setup: the recorder ships inside the DCE.
- Artifact-first doctrine — features live in documents before code; the DCE keeps the two aligned.



Requirements
- VS Code 1.90+ and git.
- An agent harness: Claude Code (recommended — extension or CLI) or any agent you can point at a file. A clipboard mode covers everything else.
Installing Claude Code
Installing the DCE also installs the Claude Code extension automatically (they ship as a pack) — for most people there is nothing to do. If you'd rather not use the extension, the Claude Code CLI works just as well:
npm install -g @anthropic-ai/claude-code
Or the native installer (no Node required) — macOS/Linux: curl -fsSL https://claude.ai/install.sh | bash · Windows (PowerShell): irm https://claude.ai/install.ps1 | iex
Then sign in once with claude in any terminal (it walks you through login with your Claude subscription). The DCE auto-detects whichever you have.
License
Installing is free, and un-activated installs run a 10-cycle trial. One key, yours forever: the DCE License is $5 at aiascent.dev/dce/buy — the streaming runs on your own Claude subscription. In VS Code, run DCE: Enter License Key and paste it once. Lost it? Recover by email.
Your work, your repos, and your cycle history always stay on your machine.