pgEdge's Starfleet gives every AI coding agent its own Postgres branch

pgEdge launched Starfleet on Monday, a Postgres cloud platform that gives every AI coding agent its own branch of a database. The branches deliberately never merge back into the source, and the product is meant to move agent-built prototypes into production without rebuilding the database layer.

Coding agents get an application running quickly, but the database setup that works during development often fails an organization's security, compliance or deployment requirements. Starfleet pairs database branching with agent tooling and deployment options that reach from pgEdge's hosted service to air-gapped on-premises installations.

One branch per agent

Running agents in parallel lets each of them try a different approach to the same problem, which gets awkward when they all write to one database, so every experiment gets its own copy-on-write branch. pgEdge says the branching does not replace Postgres' storage layer with a proprietary alternative, but it has not explained how the system works internally.

From a developer's view, a new database starts as a copy of the source and becomes an isolated environment: changes do not travel between the two, and each has its own connection details, MCP server address and bearer token, so a client configured with the source's credentials cannot connect. A branch also inherits the source's IP allowlist at creation, which cannot be changed later.

No merge back

Starfleet does not merge databases when the work is done. Schema changes go through tools such as Alembic or Flyway rather than anything built into the platform: if the code is merged, the migration files run against the source database, while test data is deleted together with the agent's database.

Cleanup is a practical necessity, since each database can carry a branch limit and billing starts as soon as a branch accepts connections and continues until deletion. Teams running many agents in parallel have a reason to automate teardown, and write commands require the agent to name a database ID explicitly.

MCP access, then production

Starfleet ships with pgEdge's Agentic AI Toolkit for Postgres, which CEO Phillip Merrick has called fully open source and free for all Postgres users. It bundles an MCP server, a RAG API built on pgvector and a PostgREST API for browser clients. The MCP server includes pgEdge SafeSession, which the company says stops a read-only agent from changing the database.

pgEdge cites IDC and Lenovo research finding that only 46% of AI prototypes reach production and that 82% of organizations need hybrid or on-premises environments. Starfleet lets developers start on hosted infrastructure and later run the same database in their own cloud or data center. Branching is documented only for the hosted tier, and Starfleet starts at $25 per month with a 14-day free trial.