Skip to content
/
Tier A A12 UI passthrough

1Code

1code · 21st-dev/1code · ★ 5.5k · last commit 2026-03-06

Primitive shape 7 total
MCP tools 7
00

Summary

1Code — Summary

1Code is an Electron desktop app and web app by the 21st.dev team that provides a Cursor-like UI for running Claude Code and Codex in git worktrees, with a Kanban board for session visualization, cloud background agents (running when your laptop sleeps), automations triggered from GitHub/Linear/Slack, a built-in git client, extended thinking visualization, plan mode with clarifying questions, chat forking, sub-agents with visual task display, and a REST API for programmatic agent control.

Problem it solved: Claude Code and Codex lack a polished visual UI — 1Code wraps them in a Cursor-like experience with diff previews, real-time tool execution display, git client, background cloud execution, and event-driven automations.

Distinctive traits: (1) Cloud background agents — agents run in cloud sandboxes with live browser previews even when your laptop is off; (2) Automations — trigger agents from GitHub PRs/issues, Linear tickets, Slack messages, or git events with @1code mention support; (3) PWA mobile interface for remote monitoring; (4) Chat forking — explore alternative approaches from any message; (5) REST API for programmatic control.

Target audience: Developers who want the Cursor-like IDE experience but for CLI-based agents (Claude Code, Codex), with enterprise automation and cloud execution capabilities.

differs_from_seeds: 1Code has the richest UI surface in this batch after SwarmClaw. Compared to all 11 seeds, 1Code is closest to kiro (the IDE-extension seed) in ambition — full IDE-level UX with agentic execution — but 1Code wraps existing CLIs rather than being a new IDE. The cloud background execution and event-driven automations (GitHub/Linear/Slack triggers) are unique among seeds and batch frameworks.

01

Overview

1Code — Overview

Origin

Created by the 21st.dev team. Apache License 2.0. Version 0.0.72. Website: 1code.dev. Actively maintained.

Philosophy

From README:

"Run coding agents the right way. Run agents locally, in worktrees, in background — without touching main branch." "UI that finally respects your code. Cursor-like UI with diff previews, built-in git client, and the ability to see changes before they land."

Core beliefs:

  1. AI agents need a visual UI that shows real-time tool execution, diffs, and git state
  2. "Never accidentally commit to main branch" — worktree isolation is default
  3. Agents should work even when your laptop is closed (cloud sandboxes)
  4. Event-driven automations eliminate manual agent invocation for repetitive workflows

Extended thinking

"Extended Thinking — Enabled by default with visual UX" Extended thinking is a first-class feature with a dedicated visual gradient UI, not just a hidden API parameter.

Sub-agents

"Custom Sub-agents — Visual task list for sub-agents in the details sidebar" Sub-agents have a dedicated visual display — not just logging.

Plan mode

"The agent asks clarifying questions, builds structured plans, and shows clean markdown preview — all before execution." Plan Mode adds a structured pre-execution gate with clarifying questions, similar to superpowers' brainstorming phase but as a visual UI feature.

Target users

Developers who want the full IDE experience (Cursor-like), enterprise teams needing GitHub/Linear/Slack automations, and users who need background cloud execution.

02

Architecture

1Code — Architecture

Distribution

  • Electron desktop app (macOS, Windows, Linux)
  • Web app (PWA) — https://1code.dev
  • REST API for programmatic control
  • Pre-built releases via subscription at 1code.dev (recommended)
  • Build from source (free)

Install (build from source)

# Prerequisites: Bun, Python 3.11, setuptools, Xcode Command Line Tools (macOS)
bun install
bun run claude:download   # Download Claude Code binary
bun run codex:download    # Download Codex binary
bun run build
bun run package:mac       # or package:win, package:linux

Directory structure

1code/
├── src/               # React renderer + app logic
├── electron/          # Electron main process
├── mcp-kanban/        # Kanban board MCP server
├── mcp-orchestrator/  # Orchestrator MCP server
├── mcp-socialdata/    # Social data MCP server
├── mcp-telegram/      # Telegram integration MCP
├── mcp-vault/         # Credential vault MCP
├── mcp-world/         # World knowledge MCP
├── mcp-x/             # X/Twitter MCP
├── openspec/          # OpenSpec integration
├── skills/            # Skills directory
├── hooks/             # Hooks configuration
├── .agents/           # Agent definitions
├── .claude/           # Claude configuration
├── scripts/           # Build scripts
└── resources/         # App resources

Technology stack

Layer Technology
Desktop Electron
Frontend React, TypeScript, Tailwind CSS
Build Bun, Electron Vite
Database Drizzle ORM (drizzle.config.ts present)
Backend Multiple MCP servers (TypeScript)

Required runtime

  • Bun (build/run)
  • Claude Code binary (bundled via bun run claude:download)
  • Codex binary (bundled via bun run codex:download)
  • Python 3.11 (build dependencies)
03

Components

1Code — Components

Bundled MCP servers

MCP Server Purpose
mcp-kanban Kanban board for visualizing agent sessions
mcp-orchestrator Agent orchestration tools
mcp-vault Credential vault
mcp-socialdata Social data integration
mcp-telegram Telegram integration
mcp-world World knowledge
mcp-x X/Twitter integration

UI features

Feature Purpose
Worktree isolation Each chat session in its own git worktree
Diff previews Real-time diff view as agent makes changes
Built-in git client Stage, commit, push, PR creation, branch management
Plan mode Clarifying questions + structured plan before execution
Extended thinking Visual gradient UX for thinking budget
Chat forking Fork sub-chat from any assistant message
Kanban board Visual session management
Sub-agent display Visual task list for sub-agent activity
MCP server management Toggle/configure/delete MCP servers in UI
Plugin marketplace Browse/install plugins
File viewer Cmd+P search, syntax highlight, image viewer
Integrated terminal Sidebar or bottom panel, Cmd+J toggle
Voice input Hold-to-talk dictation
Message queue Queue prompts while agent is working
Rollback Roll back changes from any user message bubble

Cloud/remote features

Feature Purpose
Background agents Cloud sandboxes running when laptop sleeps
Live browser previews See dev branch running in real browser
PWA Monitor background agents from phone
REST API Programmatic agent control

Automations (Pro/Max subscription required)

Trigger Description
GitHub mentions @1code in GitHub PR/issue starts agent
Linear tickets @1code in Linear starts agent
Slack messages @1code in Slack starts agent
Git events Push, PR, any git event
Conditions & filters Control when automations fire
Silent mode Background automation without response

Skills and agents (from repo structure)

  • skills/ directory — custom skills
  • .agents/ directory — agent definitions
  • hooks/ directory — hook configurations
  • openspec/ — OpenSpec integration
05

Prompts

1Code — Prompts

1Code ships skills, agents, and hooks directories but the specific content of these files is not publicly documented in the README. The framework bundles Claude Code and Codex binaries directly, suggesting it may inject its own CLAUDE.md or AGENTS.md configuration.

Plan Mode (inferred prompt behavior)

From README:

"The agent asks clarifying questions, builds structured plans, and shows clean markdown preview — all before execution."

Plan Mode implies a system prompt instructing the agent to:

  1. Ask clarifying questions before proceeding
  2. Build a structured markdown plan
  3. Wait for user approval before executing

This is functionally equivalent to superpowers' brainstorming phase but implemented as a UI mode toggle rather than an Iron Law skill.

AGENTS.md (from repo root)

The presence of AGENTS.md in the repo root suggests 1Code configures agent behavior via this file, but the content is not accessible via public API without the file path.

Sub-agent prompts

"Custom Sub-agents — Visual task list for sub-agents in the details sidebar"

Sub-agent definitions are in .agents/ directory — format not documented in README. Custom sub-agents appear to be Claude Code's native sub-agent mechanism wrapped with visual display.

Extended thinking prompt

Extended thinking is "enabled by default with visual thinking gradient UX" — this means the Anthropic API's extended thinking budget is enabled by default for Claude models, with a custom visual rendering of thinking blocks.

Prompting technique: UI-enforced gates + pass-through

1Code's key prompting innovation is UI-level gates (Plan Mode, chat forking, rollback) that enforce structured thinking without injecting Iron Laws into the system prompt. The agent behavior is shaped by user interactions in the UI, not by pre-loaded skill files.

09

Uniqueness

1Code — Uniqueness & Positioning

differs_from_seeds

1Code is the most feature-complete desktop agent UI in this batch, approaching the kiro IDE seed in ambition. Like kiro (Archetype 5 — closed IDE), 1Code provides a full developer experience layer on top of AI agent execution. Unlike kiro (proprietary IDE fork), 1Code wraps existing CLI tools (Claude Code, Codex) and is open-source. Compared to all other seeds, 1Code uniquely adds: cloud background execution (agents run while laptop sleeps), event-driven automations (@1code triggers from GitHub/Linear/Slack), REST API for programmatic control, and extended thinking visualization. The chat forking UX (explore alternatives from any message) is absent from all seeds. The combination of local worktree execution + cloud sandbox execution + PWA monitoring + REST API creates a breadth of execution contexts not found elsewhere.

Positioning

  • "Cursor-like desktop app" — explicitly targets Cursor users who want agent-first IDE experience
  • Apache 2.0: more permissive than RunMaestro's AGPL-3.0
  • Free to build from source; Pro/Max subscription for cloud features
  • 21st.dev ecosystem (the same team that makes 21st.dev component tools)

Observable failure modes

  • Cloud features (background agents, automations) require paid subscription — splits capabilities between free/paid tiers
  • "Claude Code binary download required" step: 1Code bundles Claude Code, coupling release cadence to Anthropic's CLI releases
  • Automations require @1code mentions in external platforms — friction for existing workflows
  • No git worktree isolation for cloud sandboxes (cloud-only architecture)

Inspired by

  • Cursor IDE (visual UI model)
  • GitHub Copilot (automation triggers)

Competitors named in README

None explicitly.

Cross-references

  • openspec/ directory: uses OpenSpec integration (seed framework)
  • hooks/ directory: likely Claude Code hooks
  • .agents/ directory: agent definitions
  • Multiple MCP servers bundled (mcp-kanban, mcp-orchestrator, etc.)
04

Workflow

1Code — Workflow

Interactive workflow

Phase Action Artifact
1. Plan Agent asks clarifying questions → builds structured plan Markdown plan preview
2. Review User approves/modifies plan before execution Approved plan
3. Execute Agent works in isolated worktree; real-time diff + tool display Changes in worktree
4. Review diffs Visual diff with git activity badges on messages Staged changes
5. Commit/push Built-in git client: stage, commit, push Git commit + PR
6. Fork exploration Fork sub-chat from any assistant message to explore alternatives Parallel sub-chat

Background agents workflow

Phase Action Artifact
1. Submit Create task via UI or REST API Task record
2. Cloud dispatch Agent runs in isolated cloud sandbox Background execution
3. Monitor PWA on phone or web UI shows progress Status updates
4. PR delivery Agent commits, pushes branch, opens PR automatically GitHub PR

Automation workflow

Phase Action Artifact
1. Trigger @1code mention in GitHub/Linear/Slack or git event fires Automation event
2. Filter Conditions/filters check if automation should fire Pass/skip
3. Agent creation New background agent created for this event Agent instance
4. Execution Agent runs with injected event context Changes + output
5. Response Agent posts results back to source (GitHub comment, Linear update) Comment/update

Approval gates

  • Plan Mode: user reviews plan before execution starts
  • Chat forking: user controls which forks to continue
  • Rollback: user can revert changes from any message
06

Memory Context

1Code — Memory & Context

Memory support

From README:

"Memory — CLAUDE.md and AGENTS.md support"

1Code supports the standard CLAUDE.md and AGENTS.md context injection pattern. Each agent session reads the project's CLAUDE.md for context.

Database

drizzle.config.ts is present in the repo root, indicating 1Code uses Drizzle ORM for structured data persistence. The schema is not publicly documented, but likely stores:

  • Session/conversation history
  • Agent configurations
  • Kanban task state
  • Automation definitions and execution history

Worktree state

Each session's git worktree persists until the user closes/deletes it. Changes are reviewable and rollback-capable.

Cloud execution state

Background agents run in cloud sandboxes. Their state is managed by the 1code.dev platform (Pro/Max subscription required).

Chat forking

Forked chats create separate conversation branches from any assistant message — these are persistent and navigable, enabling exploration of alternative approaches.

Message queue

Prompts queued while an agent is working are persisted and delivered when the agent becomes ready.

Context compaction

Not documented explicitly. Delegated to the underlying Claude Code/Codex runtime.

Cross-session

CLAUDE.md and AGENTS.md provide cross-session behavioral context. No explicit vector store or graph memory.

07

Orchestration

1Code — Orchestration

Multi-agent support

Yes — multiple concurrent worktree sessions, sub-agents, and automation-spawned agents.

Orchestration pattern

Parallel fan-out (default sessions) + hierarchical (sub-agents spawned from parent session).

Isolation mechanism

Git worktree — each chat session runs in its own isolated worktree. Background cloud agents run in isolated cloud sandboxes.

Sub-agents

Visual task display in sidebar for sub-agent activity. Sub-agents are likely Claude Code's native sub-agent mechanism with 1Code providing the visual display layer.

Cloud execution

Background agents run in isolated cloud sandboxes (1code.dev platform, Pro/Max required). Live browser previews of dev branches.

Event-driven automations

  • Triggers: GitHub PR/issue mentions (@1code), Linear ticket mentions, Slack messages, git events
  • Creates background agents per event
  • Agents respond back to source (GitHub comment, Linear update)
  • Fire-and-forget execution

Multi-model support

Yes — supports Claude Code (Anthropic models) and Codex (OpenAI models). Model selector in UI for switching between models/providers.

Execution mode

Interactive loop for local sessions + event-driven for automations.

REST API

POST https://1code.dev/api/v1/tasks
{
  "repository": "https://github.com/your-org/your-repo",
  "prompt": "Fix the failing CI tests"
}

Programmatic one-shot task execution. Agent runs in cloud sandbox, delivers PR.

Context compaction

Not handled at the platform level; delegated to Claude Code/Codex.

08

Ui Cli Surface

1Code — UI & CLI Surface

Desktop application

Electron desktop app — the primary local interface. Cursor-like experience.

Web app / PWA

Web version at 1code.dev for remote monitoring of background agents from phone/tablet.

No standalone CLI binary

No bin field in package.json. All interaction is through the desktop app or web UI (or REST API).

REST API

curl -X POST https://1code.dev/api/v1/tasks \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "repository": "https://github.com/your-org/your-repo",
    "prompt": "Fix the failing CI tests"
  }'
  • Async execution with status polling
  • Follow-up messages to running tasks
  • Agent commits, pushes, opens PRs automatically

Key UI features

Feature Notes
Worktree panel Each session in its own isolated worktree
Real-time diff view See agent changes as they happen
Tool execution display Live view of bash commands, file edits, web searches
Built-in git client Stage/commit/push/PR without leaving app
Git activity badges Git operations shown on agent message bubbles
Plan Mode Clarifying questions + structured plan before execution
Extended thinking Visual gradient UX for thinking blocks
Chat forking Fork sub-chat from any message
Rollback Revert changes from any message bubble
Kanban board Visual agent session management (via mcp-kanban)
Sub-agent task list Visual display of sub-agent activity in sidebar
MCP management Toggle/configure/delete MCP servers in UI
Plugin marketplace Browse/install plugins with one click
File viewer Cmd+P search, syntax highlight, image viewer
Integrated terminal Cmd+J to toggle sidebar/bottom panel
Voice input Hold-to-talk dictation
Message queue Queue prompts during active agent work

Automations UI

  • Visual execution timeline for past automation runs
  • Trigger configuration (GitHub, Linear, Slack, git events)
  • Condition/filter editor
  • Silent mode toggle

Observability

  • Real-time tool execution display
  • Automation execution timeline
  • No explicit JSONL audit log or replay capability

Related frameworks

same archetype · same primary tool · same memory type

Goose (Block/AAIF) ★ 46k
A12 UI passthrough

General-purpose AI agent (not just code) with security-first tool inspection, recipe-based shareable configurations, and 15+ LLM…

Vibe Kanban ★ 27k
A12 UI passthrough

Eliminate the overhead of planning, switching between agent terminals, and reviewing diffs by providing a single web dashboard…

Crystal (stravu) ★ 3.1k
A12 UI passthrough

Manage multiple parallel AI coding sessions in isolated git worktrees from a single desktop GUI.

Maestro (RunMaestro) ★ 3.0k
A12 UI passthrough

Orchestrate unlimited parallel AI agent sessions with a keyboard-first desktop app including Group Chat coordination and Auto Run…

AgentsMesh ★ 2.1k
A12 UI passthrough

Multi-tenant workforce platform that gives every team member a squad of AI coding agents coordinated through channels, pod…

Open Cowork ★ 1.4k
A12 UI passthrough

Makes Claude Code accessible to knowledge workers on Windows/macOS with one-click install, VM sandbox, document skills, and IM…