Rapid prototyping with AI coding agents often creates a friction point when moving from development to production. The database configurations that suffice in early stages frequently fall short of an organization's security, compliance and infrastructure requirements. pgEdge addressed this challenge by introducing Starfleet, a Postgres platform that combines database branching capabilities with agent-friendly tooling and flexible deployment options—from pgEdge's own hosted service through to isolated on-premises setups, all built on standard community Postgres.

Isolated branches for parallel agent experimentation

Running multiple coding agents simultaneously allows teams to explore different solutions to the same problem in parallel. When all agents share a single database, coordination becomes complex. Starfleet solves this by providing each agent with its own copy-on-write branch, and the company maintains that this approach avoids replacing Postgres' underlying storage layer with proprietary alternatives.

Databricks pursues a different strategy with Lakebase, which operates on Neon. That platform separates storage from compute, making branching a lightweight metadata operation with Postgres pages residing in object storage.

pgEdge has not disclosed the technical details of its branching implementation. From the user perspective, a new database branch begins as a replica of its source, then operates as a fully isolated environment. Data modifications remain confined to each branch, and connection credentials differ between them.

The isolation extends to agent tooling as well. Each environment receives its own MCP server endpoint and authentication token, preventing a client configured for the source database from accessing a branch. IP allowlists are inherited from the source at branch creation and cannot be modified afterward.

When an agent requires access from an additional address, developers must update the source database's IP allowlist and provision a new branch, which will contain the source's current data but none of the previous branch's modifications.

Git-aligned database environments and schema management

The CLI tool links a project directory to a database by storing the database or branch ID in .pgedge/link.yaml, while pgedge env pull populates the folder's .env with the DATABASE_URL. An agent developing a feature can then reach its corresponding database without manual credential distribution. Most read operations automatically target the linked database; when using a branch link, connection commands point to the branch, but write operations require the agent to specify a database ID explicitly, adding a safeguard against unintended modifications.

Starfleet does not include built-in database merging. Schema changes follow existing patterns, leveraging tools such as Alembic or Flyway rather than Starfleet-native functionality. When code merges, migration files accompany it and execute against the source database. Test data created in agent branches is discarded when those environments are deleted.

Cleanup of experimental environments carries practical weight. Each database has a branch limit, and billing commences once a branch is ready to accept connections and continues until deletion. Teams running multiple agents will likely benefit from automating this teardown process. Yugabyte is addressing similar challenges through a fleet-management platform that exposes provisioning, branching, scaling, migration and teardown operations to agents via MCP.

Developers handle schema changes the same way they already do, using tools like Alembic or Flyway rather than something built into Starfleet.

Agent access controls and safety mechanisms

Starfleet includes pgEdge's Agentic AI Toolkit for Postgres, which pgEdge CEO Phillip Merrick has characterized as fully open source and available at no cost to all Postgres users. The toolkit provides an MCP server enabling coding agents to interface directly with the database. It also offers a RAG API leveraging pgvector for retrieving data stored in Postgres, alongside a PostgREST API granting browser clients direct database connectivity—a pattern employed by Lovable and comparable application builders.

The MCP server incorporates pgEdge SafeSession, which pgEdge states prevents agents with read-only permissions from altering the database. This operates in conjunction with the distinct credentials and IP allowlists each database environment inherits.

The MCP server includes pgEdge SafeSession, which the company says prevents an agent with read-only access from making changes to the database.

Scaling from hosted prototype to on-premises deployment

pgEdge references research from IDC and Lenovo indicating that only 46% of general AI and agentic AI prototypes advance to production, and that 82% of organizations require hybrid or on-premises environments for deploying AI workloads.

Starfleet enables developers to begin work on pgEdge's hosted infrastructure and subsequently migrate the same database to private cloud or on-premises deployments, including fully air-gapped setups via pgEdge Enterprise Postgres, with the capability to expand into highly available multi-region configurations.

A current constraint exists: pgEdge's documentation presently covers branching only for the hosted tier. The company has not clarified whether agent workflows will function identically when a database transitions to a customer's own cloud or data center environment.

Starfleet pricing begins at $25 per month with a 14-day free trial.

Starfleet starts at $25 per month with a 14-day free trial.