Silo

Give agents a local database for project work.

Keep findings, plans, and progress available across agent sessions without adding working data to Git.

Use Silo when several agents need to build on the same work:

• Audit a dataset. Agents record findings in a common table. Another agent queries the findings together to identify patterns.
• Manage a migration. Plan work in a table and update progress as it proceeds. A coordinating agent uses that table to assign work to subagents.
• Continue in a later session. The next agent reads outstanding work and records what changed.

You define the tables and their rules. Silo checks writes against those rules, and agents can read the data with SQL. Agents make the decisions about what to work on; Silo stores the data they use.

Start here

Getting started walks through creating an issues table, adding a row, and reading it back. It also shows what happens when a write contains an invalid value.

Then use Design a schema to define tables for your own work.

Before you use it

• The database stays outside Git. Each repository selects a local SQLite database. A clone does not include its data.
• Write through Silo commands. SQL and saved queries are read-only.
• Sharing is optional and manual. Sharing between machines requires S3-compatible storage and explicit silo push and silo pull commands. Storage and transfer costs depend on your provider.
• There is no user-facing history of changes. The database holds your current data; Silo does not provide an audit trail.

How Silo works explains where the database lives, how the schema is enforced, and how sharing works.

Choose your next task

• Work with rows: read and change the data in a table.
• Run saved queries: save SQL and run it by name with typed arguments.
• Publish a refreshable report: turn query results into Markdown you can open in a local browser viewer. Reports run trusted JavaScript on your machine.
• Synchronize a database: share changes between machines and handle conflicts.
• Use the Tasks template: start with a bundled schema for task tracking instead of designing your own.
• Troubleshoot a problem: check a symptom and find the smallest fix.

Give your agent the instructions

Run this from any directory to read Silo's bundled agent guidance:

silo skill

It prints instructions for using Silo, with links to task guides and JSON request schemas. Agents can read those links through the CLI too:

silo skill tasks/create-table.md

For command syntax, run silo --help or a command's own help, such as silo row add --help.