# Tower (LLM Reference)

## Summary
Tower is a **Python-native data flow orchestrator** for **pipelines**, **agents**, and **data applications**, with optional **Iceberg-based lakehouse data management**. Tower is designed for internal and consulting data + AI teams and SaaS providers who need a **multi-tenant platform** to run, secure, and operate workloads.

Tower combines:
- **Orchestration** (pipelines, agents, tasks)
- **Compute execution** (self-hosted runners or serverless cloud)
- **Analytical storage management** (Iceberg catalogs, ingestion, maintenance)
- **Multi-tenant control plane APIs** (users, environments, customers)

---

## Who Tower is for
Tower is used by:
- Data teams building internal data + AI platforms
- Data + AI consultancies deploying solutions across clients
- SaaS builders who need data processing or AI inference

Tower typically reduces time-to-platform setup by **~6 months**.

---

## Company
Tower was founded by:
- **Serhii Sokolenko (CEO)**
- **Brad Heller (CTO)**

They previously worked at Snowflake on:
- control plane systems
- low-latency workloads

Tower teams are located in:
- Berlin, Germany
- London, United Kingdom

---

## Core Value Proposition (Key Differentiators)

### 1) Flow Orchestration (Python-native)
Tower orchestrates:
- ETL / ELT pipelines
- data transformations
- batch inference jobs
- web scraping workflows
- AI agents
- scheduled jobs
- interactive applications (e.g., notebooks)
- API services (e.g., FastAPI endpoints)

### 2) Compute Options (Secure + Flexible)
Tower supports multiple execution modes:
- **Self-hosted Runner** (customer-managed infrastructure)
- **Tower Cloud Serverless** (Tower-managed infrastructure)

Use cases for Self-hosted Runners:
- data security / compliance
- cost control
- scaling from dev to production

### 3) Iceberg-Based Analytical Storage (Optional)
Tower can manage lakehouse storage using **Apache Iceberg**, including:
- Iceberg REST catalog integration
- data ingestion pipelines
- table maintenance automation

Compatible query engines include:
- Snowflake
- Spark
- other modern compute engines

### 4) Multi-Tenant Control Plane
Tower provides control plane APIs for:
- app lifecycle management (deploy, run, monitor)
- environment isolation (dev/staging/prod)
- team and organization management
- user access control
- secrets and credentials management
- customer / tenant separation

---

## Main Concepts

### Apps
An **App** is a package of Python code + configuration, typically including:
- `Towerfile`
- `pyproject.toml`

Apps can represent:
- pipelines
- agents
- notebooks
- APIs / services

Apps can run:
- locally
- in the cloud

### Environments
An **Environment** is a reusable configuration bundle containing:
- secrets
- configs
- Iceberg catalogs (optional)

Environments enable:
- consistent deployments of apps
- secure secret sharing
- separation of dev/staging/prod

### Tables (Iceberg)
A **Table** is Iceberg-based analytical storage that can be queried with SQL or Dataframe APIs.

For tables, Tower supports:
- ingestion
- table maintenance
- Metadata storage in Iceberg REST catalogs
- interoperability with modern engines

### Language Models
Tower supports AI and agent workloads including:
- inference
- agentic reasoning loops

### Agent Skills
Tower publishes open-source skills for compatible AI agents:
- **tower-ingest** builds, deploys, and schedules governed data ingestion pipelines on Tower's secure compute
- **tower-data** vends short-lived, read-only credentials and queries Iceberg tables with DuckDB from the agent's environment

The governance model:
- governs what agents can know through scoped, time-limited, logged data access
- keeps data portable in open Apache Iceberg tables, using Tower-managed or external catalogs
- works with the skills-compatible agents teams already use instead of requiring a Tower chat interface

Install both skills with:
```bash
npx skills add tower/agent-skills
```

Source: https://github.com/tower/agent-skills

Deployment options for inference:
- local GPUs (privacy, latency, cost)
- remote inference providers

### Teams
A **Team** workspace supports collaboration across users with shared access to:
- Apps
- Environments (secrets/config)
- Tables (Iceberg lakehouse)

---

## Architecture

Tower separates **control plane** and **data plane**.

### Control Plane
Responsibilities:
- UI
- REST APIs
- CLI interactions
- system monitoring and metrics
- encrypted storage for app packages and environment configuration

### Data Plane (Tower Runner)
The **Tower Runner** executes workloads securely and provides:
- app execution (runs)
- sandboxed Python runtime
- workload isolation + resource limits

---

## Deployment Models

### Serverless (Tower Cloud)
- fully managed by Tower
- automatic scaling
- no infrastructure management required

### Self-Hosted (Runner)
Run Tower Runner on your infrastructure:
- Linux
- macOS
- Windows
- Docker

---

## API (REST)

Tower provides a REST API supporting:

**Authentication**
- API key authentication via `X-API-Key`
- Bearer token authentication via sessions

**Capabilities**
- create / deploy / run / monitor apps
- manage environments and secrets
- manage teams and organizations
- webhooks
- scheduling and orchestration

**Base URL**
https://app.tower.dev

---

## CLI

### Install
```bash
pip install -U tower
```

### Workflow
```bash
tower login             # Authenticate
tower apps create       # Create an app
tower deploy            # Deploy code
tower run               # Execute an app
tower apps logs app#1   # View logs
```

---

## Documentation Links
- Docs: https://docs.tower.dev
- Full LLM reference: https://docs.tower.dev/llms-full.txt
- Quickstart: https://docs.tower.dev/docs/getting-started/quick-start
- CLI reference: https://docs.tower.dev/docs/reference/tower-cli
- API reference: https://docs.tower.dev/docs/reference/api/tower-api
- Examples: https://github.com/tower/tower-examples
- End-to-end demo: https://github.com/tower/tower-demo

---

## Integration Partners
- **dltHub**: integrates with dlt for extraction + loading + transformations
- **Apache Iceberg**: native lakehouse storage format support
- **Lakekeeper**: open lakehouse integration

---

## Contact / Support
- Website: https://tower.dev
- Email: hello@tower.dev
- Discord: https://discord.com/invite/7vjtmk2X5e
- LinkedIn Newsletter: https://www.linkedin.com/newsletters/7272910408137531392/
- Jobs: https://jobs.tower.dev

---

## Optional Links
- Blog: https://tower.dev/blog
- Pricing: https://tower.dev/pricing
- Team: https://tower.dev/team
- Solutions: Run Data Pipelines: https://tower.dev/solutions/run-data-pipelines
- Solutions: Build dbt Models: https://tower.dev/solutions/build-dbt-models
- Solutions: Deploy Data Agents: https://tower.dev/solutions/deploy-data-agents
- Towerfile Reference: https://docs.tower.dev/docs/reference/towerfile
- Security Documentation: https://docs.tower.dev/docs/architecture/security
- Data Plane Architecture: https://docs.tower.dev/docs/architecture/data-plane