Skip to content
THE SUPER AGENT APP FOR YOUR IDEAS

One goal. Super agents.
Your space.

Your AI coding agents, memory and tools in one self-hosted workspace, on any device.

Get SpaceApp Try the live demo
Free to self-host ·Open source under the MIT License· Your AI accounts
Interactive Space workspace

Runs in your browser with local demo data. No account, no AI provider.

Open demo in a new tab
YOUR SPACE. THE INTERACTIVE WORKSPACE MOCK.

Start in Room 1. Switch rooms, type and explore the workspace.

YOUR COMPLETE VIBE CODING APP

Everything you need to go from “what if” to “it works.”

Why I built SpaceApp

I built it to be my main work tool. I'm a developer! And it became my main tool and work environment. I no longer need to open email, social media or any development program: absolutely nothing. I'm inside SpaceApp all day; I even watch YouTube videos and listen to music through SpaceApp.

Nikolaos Theodoropoulos, founder and CEO of SpaceApp

25 years in IT, 17 of them building and running a large e-commerce platform. He builds SpaceApp with the AI coding agents that run inside it.

SAME PROJECT. EVERYWHERE.

Your desk is optional.

Pick up the same work on your phone, tablet or computer. Your workspace stays with you.

PhoneTabletDesktopYour browser
BRING YOUR AGENTS

11 installable AI coding CLIs. One room.

Setup starts with OpenCode; add the others with one command each when you need them. What can run depends on the accounts you sign in to.

  • OpenCode
  • OpenAI Codex
  • Claude Code
  • Gemini CLI
  • Qwen Code
  • Kimi Code
  • Grok Build
  • DeepSeek CLI
  • Cursor
  • GitHub Copilot
  • Autohand
A room with four AI coding CLI panes working side by side
Four CLI panes side by side in the interactive demo, simulated in your browser.

Coming next Hermes, Oh My Pi, Qoder, Muse, Droid: declared adapters without a public installer yet.

01THINK OUT LOUD
Coming next

Talk.
SpaceApp takes action.

Have a real conversation with your workspace. Open projects, direct agents and ask what’s done. Keep your hands on the idea.

Two-way voice conversationVisual context, when you share itWorkspace control and task updates

Live voice runs on the maker's own SpaceApp today. In the public release it needs a Gemini key on the application core, which the launcher cannot set yet.

PRODUCTION LIVE AGENT & VOICE ORCHESTRATION ENGINE
  • Full-Duplex Bidirectional Audio: Real-time streaming over persistent WebSocket & WebRTC channels directly to the Gemini Multimodal Live API.
  • Low-Latency AudioWorklet & 16kHz PCM: Client-side linear PCM16 resampling and intelligent Voice Activity Detection (VAD) with barge-in interruption and echo cancellation.
  • Autonomous Workspace Action Dispatch: Bidirectional function calling protocol dispatching live actions across parallel CLI agent terminals, pane layout tiling, and room navigation.
  • Real-Time Memory & Context Sync: Continuous live transcript streaming synchronized with canonical pgvector memory ingestion and real-time context recall without prompt drift.
ACTUAL SPACEAPP LIVE PANEGEMINI LIVE
Loading SpaceApp…
Think out loud

Start a conversation with your workspace.

Example conversation · No microphone requiredExplore SpaceApp
02CONTEXT THAT STAYS

Your next idea
doesn’t start
from zero.

Decisions, discoveries and project knowledge, connected in one canonical memory you can search and browse across sessions.

“Use the approach we agreed on.”Less re-explaining. More building.

Space Memory keeps decisions and project knowledge as searchable records with a graph view. Vector search is not configurable yet.

PRODUCTION RAG & MEMORY RETRIEVAL ENGINEComing next
  • PostgreSQL + pgvector & Temporal: End-to-end RAG and retrieval engine (packages/memory-graph) backed by containerized PostgreSQL with pgvector and Temporal workflows for idempotent ingestion.
  • Deterministic 640-char Windowing: Automated credential and secret sanitization with deterministic 640-character sliding passage windowing over execution traces and technical documentation.
  • Multi-Stage Hybrid Search: Dense vector embeddings combined with sparse TF-IDF/lexical scoring and strict numeric identifier matching to preserve exact IDs and timestamps without vector drift.
  • MCP Gateway Context Feed: Model Context Protocol gateway feeding bounded, high-relevance context directly into supervised LLM agent runtimes.
ProjectsKnowledgeConnections
ACTUAL SPACEAPP 3D GRAPHCONNECTED KNOWLEDGE
Loading SpaceApp…
One connected memory

Project knowledge and decisions share one graph.

Example knowledge graph · Rendered in real timeExplore SpaceApp
COPILOTS

Copilots for the work around the code.

Coming next

Each copilot is a pane with its own CLI agent, memory and activity feed. You answer its questions and reviews in the rail; how much it does alone is your choice. They run on the maker's own SpaceApp today; the public release has the copilot panes but not yet the tools their agents work with.

  • Mail Copilot

    Works through your Gmail with gradual autonomy: it drafts and asks first, acts under rules you approve, or acts on its own when you allow it.

    Coming next

  • Job Copilot

    Works through your queue of job postings in the pane browser, asks you for what it cannot answer, and keeps a tracker of your applications.

    Coming next

  • Marketing Copilot

    Drafts posts from your briefs and publishes them on your own accounts through the pane browser after your review.

    Coming next

  • Freelance Copilot

    Finds client work on freelance platforms and writes proposals under daily limits, with each send checked first.

    Coming next
Mail Copilot with a reply waiting for review, Job Copilot with its queue and Marketing Copilot with a LinkedIn post in review
The Copilots room in the interactive demo, with invented data.
Mail Copilot on a phone: a reply to approve or reject, with tabs for the three copilots
The same room on a phone.

Copilots are optional modules: new installs start with none of them. Turn one on with optionalModules in the launcher's config.json, then run npx --yes run-spaceapp@latest repair. Copilots that use the pane browser need the large profile.

RESEARCH COUNCIL

Ask once. Get a council, not a guess.

Coming next

Several models work on the same question independently, then rank each other's answers. It runs on the maker's own SpaceApp today; the public release does not yet give the Council a model to work with.

Research Council dock showing a finished question with three ranked answers and their scores
A finished Council run in the interactive demo: the agreed Top 3, with invented data.
  1. 01

    Plan

    A coordinator model reads your question, asks what is missing and proposes who answers.

  2. 02

    Answer

    After your approval, each member answers on its own from the same evidence.

  3. 03

    Score

    The positions are merged without names and every member scores them.

  4. 04

    Agree

    The coordinator writes the agreed Top 3.

HOW IT WORKS

Three moves. One calm workspace.

Space is not another chat window. It is a system of rooms where every kind of work has its place.

  1. 01

    Create a room

    Give every mission its own room — a launch, a research thread, an ops shift — so contexts never collide.

  2. 02

    Add your panes

    Chat with agents, attach real CLI terminals and pin browser views (large profile), all side by side.

  3. 03

    Let the work flow

    Agents keep their context, artifacts stay within reach, and you stay in flow while the room does the heavy lifting.

DESIGNED AS A SYSTEM

Keep the work together, not just the chat.

Every pane in a room exists for a reason — and stays exactly where you need it.

ROOM AGENT

Give the room an agent that watches every pane.

The Room Agent watches every pane in its room and shows which ones wait for you.

Coming next Talk to it about the room, and with Auto-answer on it answers the prompts that wait for you, so the agents keep working while you are away.

Coming next: it warns when two panes change the same files, and their work joins one release train.

  1. Watch the panes
  2. Show what waits for you
Pane overviewWaiting panes
SHARED AGENT FABRIC · COMING NEXT

Different agents. One shared operating context.

CodexClaude CodeGeminiOpenCodeAutohand CodeQwen CodeKimi CodeGrok BuildDeepSeekCursorGitHub CopilotHermesOh My PiQoderMuse CodeDroid

Codex, Claude Code, Gemini, OpenCode, Autohand Code, Qwen Code, Kimi Code, Grok Build, DeepSeek, Cursor, GitHub Copilot, Hermes, Oh My Pi, Qoder, Muse Code, and Droid share one governed operating layer: the same skills library, approved MCP server catalog and canonical memory, delivered through each runtime’s supported integration path.

Canonical memory uses a Room, Project and System taxonomy with tags, provenance, lifecycle states and audited change sets. Deterministic Greek/English TF-IDF topic vectors and cosine-similarity relationships connect related knowledge.

Inside Space, an Obsidian-style knowledge graph powered by Graphology, Sigma and ForceAtlas2 provides clustered and relationship views without leaving the app. Common History preserves operational continuity across all sixteen CLI runtimes.

Shared skillsGoverned MCPCanonical memoryRoom / Project / SystemTags + provenanceTF-IDF + cosineObsidian-style graph

Rooms hold the context

Sixteen panes per room, one-click presets, and warm rooms that keep every session alive between switches.

Your CLIs, one grid

11 installable CLI runtimes, with four focused starter panes, correct branding, shared history, and runtime-aware controls.

Managed browsers, built in

Live Chrome panes with tabs, device viewports, evidence capture, and one-click operator handoff (large profile).

Clips and files

Private clipboard history and a media dock for the files your agents produce.

16PANES PER ROOM
11INSTALLABLE CLIS
6ROOM THEMES
Latest stream replayBuild in public

See Space being built, live.

Join the real build sessions, follow new features as they land, and ask questions while Space takes shape. The latest YouTube livestream appears here automatically.

Watch on YouTubeAll live streamsFollow on Twitch
Previous live sessionLatest recorded stream
Day 3 –🔥 FREE AI Coding in SpaceApp — No API Costs | Live Vibe Coding

Click anywhere to start video directly

WHAT'S NEXT

Already running. Coming to your install.

These run on the maker's own SpaceApp today. They are not in the public release yet; each card says when to expect it. What shipped so far is in the release notes.

  • Freelance Copilot

    Client work on freelance platforms, proposals under a send check.

    Coming next

  • Pane coordination

    The Room Agent warns when two panes change the same files; work joins one release train.

    Coming next

  • Media Studio

    Screenshots and short videos of your own workspace in device and social sizes.

    Coming next

  • Token usage

    Tokens per model, provider and account, with their API value.

    Coming next

  • Nous

    One mind for your whole space: it watches every room and brings you only the decisions that need you.

    Coming next

  • Space Gateway

    One gateway for your model accounts.

    Coming next

  • Android push

    SpaceApp notifications on your Android phone.

    Coming next
WHAT SHIPPED

Small releases, often.

Every release of the public installer, run-spaceapp, is on npm and in the release notes.

  1. 1.1.7

    SpaceApp and the run-spaceapp launcher move to the MIT License, OpenCode 2 is the first CLI, backup saves the database, files, memory and workflow history from one moment, a failed restore puts every database and file back, and dependency security updates.

  2. 1.1.6

    Team accounts with a CLI container per user (off until the owner runs user enable), Codex sandboxing where the host allows it, install telemetry only after the owner's Yes, and session, WebSocket Origin and framing hardening.

  3. 1.1.5

    Security hardening: other local sites cannot use your session, child processes no longer receive the session secret, and optional modules start off on new installs.

  4. 1.1.4

    CLI panes start in their own room workspace, an update notice with a Copy button, first prompts in the setup wizard, and opt-in first-answer telemetry.

  5. 1.1.3

    The first OpenCode task no longer stops at a scratch-directory permission prompt, and guided setup works on every screen size.

  6. 1.1.2

    The first free OpenCode model is checked with a real tool call, with Big Pickle tried first.

  7. 1.1.1

    The install wizard respects the telemetry answer (the default is No), and small-profile installs no longer roll back.

SECURITY AND LICENCE

Your host. Your data. Clear terms.

SpaceApp runs on your own machine. The AI providers you connect receive what you send them, under their own terms.

  • Hardened in 1.1.5

    • Other sites on your machine cannot read your session or its CSRF token.
    • /metrics answers only the host itself and administrators.
    • Agent CLIs, demo projects and browsers do not receive the session secret or the database URL.
    • Third-party images are pinned by digest, and every Admin Mode file change is audited.

  • MIT License

    Open source, copyright INTRATRADE SINGLE MEMBER P.C. Use, copy, modify, distribute and sell it, including self-hosting and commercial use, as long as the copyright and licence notice stay with it. Read the LICENSE.

  • Telemetry only with your Yes

    The install wizard lists what is sent and asks; the default is No. With Yes, a random install ID links the install, setup and first-answer events and at most one activity event a day; no files, prompts or credentials. Security & privacy guide.

YOUR NEXT IDEA STARTS HERE

Make yourself
some space.

Install once. Connect your AI.
Start building from wherever you are.

Free to self-host · Bring your own AI
Explore CLI commands & recovery guide
  1. Install Node.js 20.11+.
  2. Open a terminal as your normal (non-root) user.
  3. Run the command below. If Docker is missing, the launcher installs Docker Engine.
Install with npxnpx --yes run-spaceapp@latest install
System requirements and first setup +

Small profile: 4 CPUs, 7 GiB RAM and 6.5 GiB free disk. Browser panes need the large profile: 8 CPUs, 15 GiB RAM and 11 GiB free disk. Node.js 20.11+ is required; the launcher installs Docker if it is missing. OpenCode is the first CLI in setup.

To continue from another device, connect to the same reachable host. Keep that host running. Phones and tablets access your workspace; they do not run the Docker installer.

CLI REFERENCE & OPERATIONAL GUIDE

Every command you need to install, recover, and operate.

Run directly in Command Prompt, Terminal or Bash with npx — no global installation required. First-time account registration, password resets, automatic health checks, and lifecycle management.

Default closed · Click to expand or select any category above
A FEW THINGS TO KNOW

Before you
make space.

What is SpaceApp?

SpaceApp is a super agent app for building with AI. Start with a conversation, then bring your agents, tools, files, previews, plans and shared memory into the same space. Use as much of it as you need.

Can I keep working from my phone or tablet?

Yes, once your installation is reachable from them. By default SpaceApp listens only on its own host; to use it from a phone, tablet or another computer, give it a reachable address with TLS and a firewall (see Remote & mobile access in the docs). The host must stay on.

Is SpaceApp free?

SpaceApp is free to self-host. Bring your own AI accounts or API keys; provider subscriptions and usage are separate.

What licence does SpaceApp use?

SpaceApp is open source under the MIT License, copyright INTRATRADE SINGLE MEMBER P.C. You may use, copy, modify, distribute and sell it, including self-hosting and commercial use, as long as the copyright and licence notice stay with it. The LICENSE file on GitHub has the full terms. Read the LICENSE ↗

Do I need to install every AI tool?

No. Start with one conversation or one agent, and add providers and accounts as your work grows. The setup starts with OpenCode; you choose what to connect next.

Are the copilots on when I install?

No. The copilot panes are optional modules, and new installs start with none of them; you turn one on with optionalModules in the launcher's config.json and npx --yes run-spaceapp@latest repair. Their agents' tools are not part of the public release yet: the copilots work on the maker's own SpaceApp today.

Which features are in the public release?

Everything on this page without a label is in the current public release. A card marked “Coming in” a version arrives with that release; a card marked “Coming next” runs on the maker's own SpaceApp and is not public yet.

Where does my work live?

Your workspace runs on your own installation. Connected AI and external services process the information you send to them under their own policies.

Does SpaceApp send telemetry?

Only if you say yes. The install wizard lists what is sent and asks; the default is No. With Yes, a new install gets a random install ID that links its events: the install (OS, CPU, RAM, profile, versions), owner setup, the first CLI and the first answer, and at most one activity event a day (screen size, browser, UI language, version). No files, prompts, answers or credentials are sent, and the IP address is not stored. Nothing is sent with No.

Can I try it before installing?

Yes. The live demo runs in your browser with local demo data and no account; it does not contact AI providers or a real installation. When you like it, install your own workspace.

FOR TEAMS

Bring SpaceApp
to your team.

A paid pilot for software and AI teams of 3–15 developers who already use AI coding CLIs. We install SpaceApp for each developer on their own machine, using your own API keys.

One install per developer, each the owner of their own SpaceApp OpenCode and Claude Code on your own API keys, with your MCP servers Nightly backup with a tested restore 2 workflows or MCP integrations built with you Team training and 30 days of support 2–4 weeks · from €1,500, fixed quote, 50% deposit

CLIs in the pilot: OpenCode and Claude Code. Other CLIs are added when they pass a task on your install during setup.

Book a paid pilot Pilot one-pager (PDF)

How the pilot works

  1. A 30-minute scoping call: your team, your tools and the two workflows.
  2. A written fixed quote and statement of work; 50% deposit.
  3. Week 1: install and configuration on each developer's machine.
  4. Weeks 1–3: the two workflows or MCP integrations.
  5. Last week: team training.
  6. Handover with a written runbook; 30 days of support start.
  7. A review call: continue or stop. No lock-in: the install stays yours.

What the pilot is not

  • Not a production security certification. SpaceApp is actively developed software with a public list of open findings.
  • Not tenant isolation. Each install has one trusted owner, and rooms organise work rather than separating untrusted users.
  • Not a sandbox for the host. Agents work with the folders and credentials an install is given. Scope them per machine.
  • Not hardened for the open Internet. If you expose an install, TLS, proxy access, firewall and log retention are yours.
  • Production use needs at least a threat model for your environment, monitoring of the nightly backups and a review of the logs against your requirements. These can be scoped as separate work.

MIT licensed: no licence fee. See the security boundaries.

GET IN TOUCH

Tell us what
you’re building.

Questions, feedback or the workspace your team needs. We’ll reply personally.

Personal reply· Direct to the team· Privacy-first

Prefer live community chat?

Join Discord

We use these details only to reply to your message.

SpaceApp is a product of INTRATRADE SINGLE MEMBER P.C., Patras, Greece · GEMI 180769616000 · VAT EL802678365