Nexura Postgres

PostgreSQL run/session storage and a pgvector vector store for nexura-runtime.

Install

pip install nexura-postgres

Alpha (0.4): pip install nexura-postgres. See package installation.

Minimal example

This example constructs a repository without connecting. Call await repository.init_db() against a running PostgreSQL server to create the run/session/event tables. pgvector needs the vector extension. SQLite needs no add-on; without this package the runtime factory raises MissingDependencyError naming nexura-postgres.

from nexura_runtime.storage import StorageRepository, get_storage_repository

from nexura_postgres import PostgresRepository

repository = get_storage_repository("postgresql://nexura:secret@localhost:5432/nexura")
assert isinstance(repository, PostgresRepository)
assert isinstance(repository, StorageRepository)  # same protocol as SQLiteRepository
print(type(repository).__name__, repository.dsn)

Durable workflows and agents

PostgresWorkflowStore(dsn) is the workflow checkpoint store for WorkflowEngine on several hosts (revisioned checkpoints, history, single-use approvals and the execution journal on the same pool); nexura-runtime serve --workflow-state-url postgresql://... uses it for the service, together with PostgresTriggerStore(dsn), the trigger job queue (webhooks, receipts, schedules and job leases on the database clock). PostgresCheckpointStore(dsn) does the same for SDK agent runs. See Durable execution.

Status and limits

Status: alpha (0.4), Python 3.12 (the same as nexura-runtime). Tested against a local disposable PostgreSQL service with pgvector; never against a managed cloud database.

Documentation and licence

Main documentation · Full guide. MIT, Copyright (c) 2026 Liceron. See LICENSE.