The full-stack platform tuned for AI agents.

Ask an AI agent to build something real - an app, a dashboard, an API, an agent - and it does one of two things: rebuilds the backend from scratch as fragile one-off code, or stitches together a dozen third parties, each with its own API keys, billing, dashboards, and SDK to learn. Both burn tokens, take more turns, and ship more bugs.

Gipity is one full stack, and every layer is tuned for the agent. Low-level primitives - database, storage, CDN, auth, functions, deploy. Kits for the harder parts - multiplayer, vision, media - with no server setup. The most popular third-party APIs are built in, no keys to wire.

It's not a walled garden. Gipity's agent-tuned tools are the easy path most builders will want - but you're never boxed in. Bring your own API key for any integration, point at your own AWS, run whatever you want. Great defaults, real exits.

The result: fewer tokens, fewer turns, fewer bugs. And Gipity runs what it builds - hosting, monitoring, rollback included. No IDE, no code to read. Use Gipity in your browser, or pair it with your coding agent - Claude Code, Codex, or Grok - in one command.

This is the capability map. For the mission and principles behind Gipity, see about-gipity.

Build Apps & Websites

Tell the agent what you want and it builds it - HTML, CSS, JavaScript, deployed to a live URL. No IDE, no terminal, no Git.

Apps deploy to dev.gipity.ai (testing) or app.gipity.ai (production). Auto-deploy mode pushes changes live on every file edit. Load web-app-basics for best practices, app-development for backend APIs.

3D World

3D World is the 3D multiplayer game template on Gipity. All 3D World games share the same visual style, physics engine (Rapier), and multiplayer backend (Gipity Realtime). All files are fully editable.

Add a 3D World project with add name=3d-world (web agent) or gipity add 3d-world (CLI). This creates a playable 3D game with Three.js + Rapier physics + Gipity Realtime multiplayer. Key files: config.js (metadata), settings.js (tunable values), strings.js (display text), objects.js (entity factories), game.js (orchestrator), plus engine files (core.js, world.js, physics.js, etc.).

Genres: obby/parkour, tycoon, simulator, PvP combat, shooter, tower defense, horror, racing, RPG, social.

Features: Opt-in gameplay modules enabled via config.features. Available: rocket-launcher (projectile weapon with physics explosions). Example: features: { 'rocket-launcher': true } in config.js. Features auto-initialize during boot.

Regular game requests ("make a wordle", "build a quiz") should use the standard web template - they don't need the 3D template.

Load 3d-engine for the engine API and a blank-slate template, or 3d-world for the playable starter, genre recipes, and multiplayer patterns.

Backend Services - No API Keys Needed

Your deployed apps can call AI models, generate images, create speech, and play sound effects through built-in API services - no third-party accounts or API keys required.

Two billing modes: you pay (simple) or your app users pay from their own credits. Load app-llm, app-image, app-video, app-tts, app-audio, or app-files for integration details.

Databases

Per-project PostgreSQL schemas. Full SQL - create tables, query data, build APIs on top. The agent handles the SQL for you.

Database count is per-plan (call credits_products to see your limit). Load app-development for functions, database setup, and API patterns.

Code Execution

Sandboxed Docker containers with JavaScript (Node 20), Python 3, Bash, and 60+ pre-installed CLI tools. Data analysis, file conversion, image processing, document generation - all without leaving chat.

Load sandbox-tools for the full tool and library list.

Automate Anything

Multi-step workflows triggered by schedule, webhook, or manual run. Each step is an AI call with full tool access.

Human-in-the-loop approval steps for anything that needs sign-off before proceeding. Load workflow for cron syntax and patterns, approvals for approval flows.

Generate Media

Images, speech, sound effects, and music - generated on demand and saved to your files.

Load tts for voice and audio details.

Search the Web & Social Media

Real-time web search and Twitter/X search (last 7 days). The agent can research current events, find documentation, check prices, and monitor social media. Load web-search for operators and tips.

Connect External Services

Gmail, Google Calendar, Slack, GitHub, Todoist, Notion, Telegram, Discord - connect accounts and the agent can read, send, and manage on your behalf. Load google-services, service-call, telegram-connector, or discord-connector for setup.

Persistent Memory

The agent remembers what matters across conversations. Two scopes:

You can ask the agent to remember or forget anything. Memory is also auto-extracted from conversations.

File Management & Recovery

Per-project file workspaces with full CRUD. Files generated by tools (images, audio, video, sandbox outputs) are saved as project files automatically. Every file write is tracked with immutable version history - you can view versions, restore individual files, or roll back entire directories to any point in time. Load version-history for recovery options.

Multi-Model AI

Claude (Sonnet, Opus, Haiku) and GPT (GPT-4.1, GPT-5) families. Switch models per agent. Extended thinking mode for complex reasoning. Cross-model queries to get second opinions. Load query-llm for cross-model queries.

User Authentication for Apps

Apps on Gipity can authenticate Gipity users with zero setup - session cookies, no OAuth configuration needed. Load app-auth for implementation details.

Real-Time Multiplayer

WebSocket rooms for games and collaborative apps. Relay (message broker) or state (server-authoritative) modes. Up to 200 clients per room. Load app-realtime for room types and client code.

Relays: drive your coding agent from the web CLI

A relay is what runs your coding agent (Claude Code, Codex, or Grok) for your web-CLI chats. Two kinds:

Then in the web CLI start a chat with /relay (or /chat claude) - your prompt switches to claude> and each message dispatches to the relay, runs your coding agent (gipity build -p "<msg>"), and streams the conversation back into the terminal view. Every prompt, tool call, and assistant turn is recorded to your Gipity project - from Claude Code, Codex, and Grok alike - and viewable live in the web client. exit returns to the normal Gipity prompt. Check pairing state anytime with gipity relay status; gipity relay revoke unpairs and removes the login service.

Switch the model for a relay chat with /model <haiku|sonnet|opus> (or any catalog id) right from the web CLI claude> prompt - the choice is stored per-chat and forwarded to your local coding agent as --model on your next message. Relay chats run on your own agent subscription, so there are no Gipity credits involved. Locally, gipity build --model <alias|id> works the same way - the flag passes straight through to the underlying agent binary (it is NOT a Gipity-only flag), so gipity build --model opus is valid and launches the session on Opus.

Credits & Billing

Gipity runs on credits. Use this to answer the common "wait, am I going to get charged / will everything break?" worry plainly:

Platform Limits

Storage, database count, concurrent chats, and deploy rate are all per-plan. Run the credits_products tool to show the user's plan and what it includes.

Where to Start

Just ask. "Build me a website." "Analyze this data." "Set up a daily email." "What can you help me with?" The agent will figure out the tools, the code, and the deployment. You describe what you want - the AI handles the how.