Tower is where AI agents securely work with your enterprise data. This repository has the Agent Skills that teach any skills-capable agent (Claude Code, Cursor, Tower Control, and others) how to use Tower.
npx skills add tower/agent-skills
- Any agent can safely access data on demand. Federate data from across your company into Tower, then delegate access to agents anywhere in the company. Tower vends short-lived, scoped credentials, so an agent gets exactly the access it needs and nothing outlives the session.
- Data infrastructure that improves itself. Agents use Tower's sandboxed storage and compute to build and maintain pipelines, notebooks, and reports without putting the rest of your data estate at risk. When an experiment works, promoting it to production is one step.
- Scale to thousands of agents on demand. Tower's serverless platform runs data jobs where your data lives, across six regions in the US, Europe, and APAC, and you only pay for what you or your agents use.
Every interaction is auditable, and Tower is enterprise ready with SOC 2, GDPR, and HIPAA compliance. If data can't leave your network, you can even bring your own compute.
| Skill | What it does |
|---|---|
tower-integration |
Gets data into Tower. Builds ingestion pipelines that land raw data from APIs, databases, SaaS tools, and files into Iceberg tables, then deploys and schedules them on Tower's managed compute. Credentials live in Tower secrets and never in code, config, or chat. |
tower-ontology |
Works out what your data means. Profiles every table, proves the grain and join keys with SQL instead of guessing from column names, and records the entity model as catalog knowledge, so the next agent starts with the semantics instead of re-deriving them. Read-only: it writes knowledge, never data. |
tower-data |
Gets answers out of Tower. Uses tower catalogs query, which vends a short-lived, read-only credential and runs SQL through DuckDB in one command, and answers questions with SQL you can audit. Reads and records data semantics with tower catalogs knowledge, so definitions carry across sessions. The agent never touches the source systems. |
Together they cover the full loop: source → Tower → understanding → answer. Tower sits in the middle, so every run has a status, history, and logs, and access is scoped and auditable. The same skills power Tower Control, Tower's first-party harness, so you can start building data agents in the browser with nothing to install.
You'll need a Tower account and the CLI:
pip install tower
tower login
npx skills add tower/agent-skillsThen ask your agent for what you want ("get our Stripe data into Tower," "how many teams ran apps last week?") and the right skill takes it from there.
Useful variants:
npx skills add tower/agent-skills --skill tower-data # just one skill
npx skills add tower/agent-skills -a claude-code # target a specific agent
npx skills add tower/agent-skills --list # see what's here firstIssues and pull requests are welcome. These skills are meant to improve through real use. See CONTRIBUTING.md for how to add a skill and AGENTS.md for the design principles every data skill here follows.
- Learn more about Tower
- Tower documentation
- Make Every Agent a Data Agent, the announcement post