What's Inside SINAS
Three core systems: AI agents, Python functions, and state management
Agents Call Agents
Build complex AI systems by composing agents. One agent handles routing, another does research, a third writes code. Each agent is a tool the others can use. Chain them together to solve complex problems.
Skills System
Reusable instruction modules that agents retrieve on-demand. Progressive disclosure: agents call skills when needed. Or preload them into system prompts. Share expertise across all your agents.
Rich Tool Ecosystem
Agents can use Python functions, MCP tools, other agents, and skills—all as tools. Functions run in isolated containers with full authentication context. Execution tracking shows exactly what happened.
Function Execution
Python functions run in isolated Docker containers. Receive webhooks, schedule with cron, or trigger from agents. Automatic execution tracking builds visual trees of nested calls.
Multi-Provider LLM
Works with OpenAI, Anthropic, or self-hosted models (Ollama, vLLM, etc). Switch providers per agent or per request. No vendor lock-in. Your infrastructure, your choice.
Built-in RBAC
Namespace-based permissions with hierarchical scopes (:own/:group/:all). Control who can create/read/update agents, functions, and state. Passwordless OTP auth. Multi-tenant ready.
GitOps Ready
Declarative YAML configuration for agents, functions, skills, and schedules. Validate before deploying. Track changes in git. Idempotent updates. Infrastructure-as-code done right.
Management Console
Web UI for configuring agents, viewing execution logs, managing users and permissions. Monitor workflows, debug execution trees, test functions—all from one interface.
Simple Architecture
FastAPI backend, PostgreSQL database, Docker for functions, optional ClickHouse for analytics.
🐳 Docker Compose Stack:
├── FastAPI (Python backend)
├── PostgreSQL (main database)
├── Docker (for function containers)
├── Management Console (React)
├── ClickHouse (optional analytics)
└── Redis (optional caching)
Functions run in isolated Docker containers. ClickHouse and Redis are optional.
Deploy Anywhere
- Run locally with Docker Compose
- Deploy to VPS or cloud provider
- Bring your own Kubernetes charts
- No special infrastructure needed
Organize with Namespaces
- Use namespaces to separate projects
- RBAC controls who accesses what
- All projects on one deployment
- Or run multiple deployments
Get Started
Clone, configure, deploy. Running in 5 minutes.