Aster is a self-hosted, enterprise-ready, open-source AI agent platform. Run powerful AI agents, orchestrate complex multi-agent workflows, and securely connect models to your local and external tools - entirely on your own infrastructure, without compromising data privacy or compliance.
GitHub Repository · Documentation
Every time your team copies a spreadsheet, a contract, or a proprietary document into a public cloud AI tool (like ChatGPT or Claude), your data leaves your company walls.
- Data Privacy: Your intellectual property is processed on external servers and may be used for model training.
- Compliance Risks: Uploading customer data to the cloud violates strict compliance regulations like GDPR, HIPAA, and CCPA.
- Security Exposure: Running custom scripts or integrations (plugins) on your internal networks directly linked to the LLM exposes you to prompt injection and unauthorized command execution.
Aster is a self-hosted platform that runs entirely on your own hardware or private cloud. It keeps the power of state-of-the-art AI while ensuring your data never exits your secure boundary.
Through its MCP-native architecture, Aster isolates every tool and script in separate, restricted Docker containers, making it the most secure way to connect AI models to your databases, local filesystem, and internal APIs.
| Feature / Metric | Public Cloud AIs (e.g., ChatGPT Team) | General AI Toolkits (e.g., Flowise, Langflow) | Aster (Self-Hosted) |
|---|---|---|---|
| Data Sovereignty | No (Data sent to third-party cloud) | Depends on hosting | True Sovereignty. Runs on-premise. |
| Tool Execution Security | None (Running arbitrary code is unsafe) | Process runs directly on host machine | Strict Sandbox. Rootless Docker containers with egress allowlists. |
| Privacy Isolation | No (Admins can see user logs) | Basic or no database-level tenant isolation | Strict Privacy. Row-Level Security (RLS); even admins cannot read private chats. |
| Agent Autonomy | Limited to custom GPT instructions | Linear pipelines | A2A Coordination. Master Planner delegates subtasks to specialist sub-agents up to 5 levels deep. |
| Capabilities Extension | Locked to app store or plugins | Complex coding required | Skill Creator. Generates and validates new capabilities dynamically using natural language. |
Aster divides processing, execution, and data storage into distinct layers.
- The Frontend WebUI renders the chat and designer.
- The Backend Host coordinates agent logic, databases, and cron schedules.
- The Orchestrator manages short-lived, sandboxed Docker containers running MCP (Model Context Protocol) servers.
graph TD
User(["User Client"]) -->|HTTPS| WebUI["WebUI (React SPA)"]
WebUI -->|REST / SSE| Host["Host Server (Node.js / Fastify)"]
Host -->|Row-Level Security Guard| DB[("PostgreSQL + pgvector")]
subgraph Storage & Context
DB -->|RAG| Context["Long-term Memory / Vector Store"]
Context -->|Inject| Host
end
subgraph AI Coordination
Host -->|Orchestrate| Planner["Master Planner Agent"]
Planner -->|API Request| LLM["LLM Providers (Claude, GPT, Ollama)"]
LLM -->|Response| Planner
Planner -->|A2A Delegation| SubAgent["Specialist Agent"]
end
subgraph Safe Execution Sandbox
SubAgent -->|Call Tool| MCP["MCP Orchestrator"]
Planner -->|Call Tool| MCP
MCP -->|Spawn/Access| Docker["Rootless Docker Daemon"]
Docker -->|Isolate| Env1["MCP Server: Local Filesystem"]
Docker -->|Isolate| Env2["MCP Server: Search Engine"]
end
Env1 -->|JSON Output| Host
Env2 -->|JSON Output| Host
sequenceDiagram
autonumber
actor User as User Client
participant Host as Fastify Host
participant DB as Postgres (pgvector)
participant LLM as AI Model (LLM)
participant MCP as MCP Sandbox (Docker)
User->>Host: Submit prompt: "Summarize budget.pdf and search web for market trend"
Host->>DB: Query User Memory (Semantic Search)
DB-->>Host: Return matching memories / context
Host->>LLM: Send context + prompt
Note over LLM: Master Planner decides to<br/>split task into sub-actions
LLM-->>Host: Action 1: Read filesystem (budget.pdf)<br/>Action 2: Execute web search (market trend)
Host->>MCP: Spawn isolated Filesystem MCP Container
MCP-->>Host: File contents (raw text)
Host->>MCP: Spawn Web Search Container (with strict outbound allowlist)
MCP-->>Host: Search results
Host->>LLM: Feed tool outputs to LLM
LLM-->>Host: Final summarized markdown report
Host->>DB: Save interaction & new learnings (Memory)
Host-->>User: Render markdown response + step traces
- Visual Chain Designer: Design multi-agent workflows graphically without writing a line of code.
- Agent-to-Agent Delegation (A2A): Break down large problems. A planning agent orchestrates and assigns sub-problems to specialist agents (e.g. database specialist, coder, researcher).
- Docker-Isolated MCP Tools: Native support for the Model Context Protocol. MCP servers are automatically run in rootless Docker containers with strict CPU, memory, and network allowlist protections.
- Long-Term Vector Memory (RAG): Auto-indexing of conversation context and text documents into Postgres
pgvectortables, split cleanly by namespace and protected by database-level Row Level Security. - Agent Skills (Open Standard): Compatible with the
agentskills.iostandard. Write reusable scripts, or use the built-in Skill Creator to construct, lint, and validate new skills interactively directly inside the chat. - Automated Cron Scheduling: Run agent tasks or chains on a schedule (e.g. daily reports, system health checks) with optional chat continuation.
- Strict Privacy Policy: Hardcoded database separation ensures that administrators can never view user chats or modify private memory without explicit user opt-in.
- Frontend: React, TypeScript, Vite, Tailwind CSS, shadcn/ui
- Backend: Node.js, Fastify, TypeScript
- Database: PostgreSQL with
pgvectorextension - Database Migrations: Flyway (Community Edition)
- Containerization: Docker, Docker Compose (configured for rootless deployments)
- Protocols: Model Context Protocol (MCP), SSE (Server-Sent Events)
| Component | Minimum | Recommended | Notes |
|---|---|---|---|
| Operating System | Linux / macOS | Ubuntu 22.04+ / macOS | Windows supported via WSL2 |
| Docker | Engine 24+ | Engine 26+ | Rootless mode strongly recommended |
| Docker Compose | v2.20+ | v2.26+ | Run as docker compose (not legacy docker-compose) |
| RAM | 2 GB | 4 GB+ | Increase if running local embedding models |
| Disk Space | 5 GB free | 10 GB+ | Required for Docker images and caching |
| Network Tools | openssl, curl, jq | latest | Utilized by installer scripts |
Note: At least one AI Provider API Key (e.g., Anthropic, OpenAI, or a local Ollama instance) is required to operate.
Run the interactive setup script on your target server. It configures the Docker environment, generates secure database credentials and session secrets, sets up database migrations, and prompts you to create the initial admin account:
curl -fsSL https://raw.githubusercontent.com/daksh777f/Aster/main/scripts/install.sh | bashTo customize your setup manually, clone the repository, copy the example environment file, customize your secrets, and launch the stack:
# Clone the repository
git clone https://github.com/daksh777f/Aster.git
cd Aster
# Set up environment variables
cp .env.example .env
# Edit .env and change FLYWAY_PASSWORD, ASTER_APP_PASSWORD, and session secrets
# Start the containers
docker compose up -dOnce running, visit http://localhost:5173 in your browser to log in and start using Aster.
For step-by-step guides, configuration settings, and administrator consoles, please refer to the files in the local docs directory:
| Topic | English Manual |
|---|---|
| Introduction | EN |
| Installation Guide | EN |
| Environment Variables Reference | EN |
| Agent Concepts | EN |
| Custom Agent Skills | EN |
| Visual Chain Engine | EN |
| Vector Memory & RAG | EN |
| Security Architecture | EN |
| Cron Automations | EN |
| API Endpoints | EN |
Aster is dual-licensed:
- Open Source: GNU Affero General Public License v3.0 - free for personal use, self-hosted deployments, and open-source projects.
- Commercial: Commercial License - for organizations wishing to use Aster without AGPL copyleft obligations (e.g. inside proprietary products or as white-label offerings).
For licensing inquiries or community help:
- Repository: https://github.com/daksh777f/Aster
Copyright (c) 2026 Daksh Goel - https://github.com/daksh777f/. All rights reserved.