pgvector-index-manager
A command-line tool for keeping your pgvector indexes in line with a configuration you control. It checks what exists, shows you what needs changing, and can build and replace HNSW indexes when you apply those changes.
DownloadVersion 0.1.1pgvector-index-manager
About pgvector-index-manager
Having an index is one thing. Knowing it still has the right settings, covers the right data and finished building properly takes a little more attention. That gets more interesting when you have several indexes for different parts of the same table.
pgvector-index-manager compares the database with your configuration. It checks the vector dimensions, distance metric, HNSW settings, filters and index health, then works out what needs to happen.
The workflow starts with inspection. You can check the current state, review a plan or do a dry run before applying anything. The inspection commands can use read-only credentials; applying changes needs permission to manage indexes.
When a replacement is needed, it builds the new index concurrently alongside the existing one and verifies its structure before giving it the expected name. The previous managed index is then retired. You can set a grace period before that cleanup, and the remaining wait survives a restart.
It also keeps track of which indexes belong to which configuration. That matters when different configurations manage different subsets of the same table. A job looking after one group of documents should stay within its own set of indexes. Conflicting indexes outside its ownership are protected by default.
You can run it once when you need a change, or explicitly enable continuous mode to keep checking and correcting differences from the configuration.
It’s written in Go, with a Docker demo for trying the workflow on synthetic data. It also sits quite naturally alongside pgvector-filterbench: Filterbench helps you measure how your searches behave, while Index Manager handles the indexes you choose to maintain.
Release history
Version 0.1.1
Fixed
- Persist retirement grace deadlines in structured index metadata so a restart cannot bypass the remaining grace period.
- Refuse to replace a same-name index owned by another managed family or ownership tag, even when
drop_unownedis enabled. - Verify the catalog structure of managed indexes, including the key layout, vector source/cast, stored type and dimensions, operator class, HNSW parameters, referenced columns, and an attested fingerprint of the complete expression and predicate.
- Detect schema namespace conflicts before building a replacement when the desired index name is already occupied by another index, table, view, materialized view, sequence, or foreign table.
- Apply configured statement and lock timeouts when recovering retirement state from an earlier release.
Changed
- Ownership metadata version 2 records the verified structural fingerprint and, for retiring indexes, an absolute UTC retirement deadline.
- Indexes carrying
0.1.0ownership metadata are rebuilt once so they acquire the stronger structural attestation. - Raise the build requirement from Go 1.21 to patched Go 1.25.12 and update pgx to the release that fixes GO-2026-5004.
Tests
- Add PostgreSQL/pgvector integration coverage for durable grace recovery, cross-family ownership conflicts, copied-comment structural drift, and namespace-conflict preflight.
- Add GitHub Actions checks for unit tests, the race detector, vet, builds, Staticcheck, Govulncheck, and PostgreSQL/pgvector 16, 17, and 18 integration tests.
External download
Download 0.1.1
Version 0.1.0
- Initial release of pgvector-index-manager.
- Go CLI with
status,plan,apply, andversioncommands. - Catalog inspection for pgvector HNSW indexes on target tables.
- Drift detection comparing current indexes against desired configuration.
- Concurrent index builds with unique temporary names.
- Ownership tracking via index definition comments.
- Advisory lock serialization across multiple manager instances.
- Verification of index health (valid, ready, live) before retirement.
- Protection against dropping unowned indexes by default.
- Configurable HNSW parameters (m, ef_construction).
- Support for vector and halfvec types with cosine, l2, and ip metrics.
- Configurable build timeouts and grace periods.
- One-shot and continuous reconciliation modes.
- Privacy-safe JSON reports with redacted connection strings and 0600 file permissions.
- Strict YAML configuration validation with helpful error messages.
- Docker Compose demo with PostgreSQL and pgvector.
- Synthetic dataset with 10,000 documents at multiple filter selectivity levels.
- Focused tests for configuration, catalog inspection, drift detection, advisory locks, and reporting.
- Go 1.21-compatible module graph and Docker build.
- Apache-2.0 license.
External download
Download 0.1.0
More software
IndexLane Safe WebP Queue
WordPress Version 1.0.1
A WordPress plugin that creates WebP copies of your images while keeping the originals. Convert on the server or use the optional browser mode when your host cannot write WebP, with progress reporting and checks before converted files are saved.
Classifieds & Advertising Platform — Go Rebuild
Go Version 1.17.10
An ongoing Go rebuild of my Laravel classifieds platform, with a React dashboard, incremental listing updates and background jobs handled through NATS JetStream. It preserves the existing platform’s behaviour while changing how the application processes and serves its data.
pgvector-filterbench
pgvector-filterbench
Go Version 0.1.2
A command-line tool for checking how pgvector search behaves once you add filters. It measures speed, compares approximate results with exact search, and checks query plans to see whether PostgreSQL is using your HNSW index.
wp-migrate-core
wp-migrate-core
Typescript Version 0.3.0
A command-line tool for getting a WordPress-to-Astro migration started. It reads a WordPress XML export, inventories content, images and links, and can generate an Astro project with a clear list of what still needs rebuilding.
hueframe
TypeScript Version 1.0.3
A tool for turning a photo into a website color theme. Drop in an image, see its colors applied to a page, adjust the result and export it as CSS variables or JSON design tokens.
SSRWire
Typescript Version 0.5.0
A command-line tool for inspecting the HTML your server sends, including content and metadata that arrive in a stream. It records when important elements appear and compares responses for different browser, search-crawler and social-preview user agents.
routeplay
routeplay
TypeScript Version 0.4.0
A command-line tool for checking that a page delivers the right content and metadata however you reach it. It compares the HTML your server sends, a fresh browser load and the result of clicking through from another page.
ReachLog
Typescript Version 0.2.1
ReachLog is a small outreach and contact manager for keeping track of who you’ve contacted, what you’ve discussed and when to follow up. It brings contacts, companies and email conversations together in a local application that works with your existing email account.
IndexLane Redirect & Internal Link Auditor
Wordpress Version 0.7.0
A WordPress plugin for finding broken internal links, redirects and old URLs left behind after a migration. It shows you where those links are used and takes you to the place where you can edit them.
routelint
routelint
TypeScript Version 0.3.1
A command-line tool for finding broken links, incorrect redirects and indexing conflicts on a live or preview website. It discovers URLs, checks what your server returns and gives you findings tied to specific pages, with the evidence behind them.
Data-Driven Publishing Platform
Data-Driven Publishing Platform
Laravel Version 0.14.1
A publishing platform for running several websites from structured source data. It handles imports, page generation, scheduled updates and search monitoring through one administration panel. Four domains are live; getting traffic to them remains the unfinished part.
Classifieds & Advertising Platform
Laravel Version 1.3.6
A classifieds and advertising platform I’ve been building and maintaining for roughly two years. It combines location-based browsing, advertiser dashboards, paid listings, invoicing, live chat and an administration panel for running the site.
Private Organic Search & LLM Activity Tracker
Elixir Version 0.39.19
A private analytics tool for tracking website traffic, organic search performance, and activity from search engines and LLM crawlers. It brings ordinary visitor statistics, crawler requests, Google Search Console and Bing Webmaster Tools into one dashboard.
ILCrawler
Go Version 0.41.0
ILCrawler is the technical SEO crawler I use for my own websites and client audits. It brings page checks, links, screenshots, performance samples and reporting into one place. It’s also the project that repeatedly convinced me I was only one rewrite away from being happy with it.
crawlreplay
crawlreplay
Rust Version 0.1.0
A command-line tool for testing robots.txt changes and new deployments against URLs found in your server logs. It helps you catch newly blocked pages, broken responses and unexpected indexing changes before putting those changes live.
Data-Driven Publishing Platform — Go Rebuild
Data-Driven Publishing Platform — Go Rebuild
Go Version 0.226.2
An unlaunched Go rebuild of my publishing platform, using PostgreSQL for application data and background jobs, React for administration and Astro for public websites. It includes source processing, publishing checks and migration tools for the existing sites.
IndexLane Crawl Fetch Inspector
WordPress Version 0.2.0
A WordPress plugin for checking redirects, indexing rules, sitemaps and structured data from your dashboard. Particularly handy when you’re launching a site or moving one.
IndexLane Asset Weight Snapshot
WordPress Version 0.2.0
A WordPress plugin for checking the CSS and JavaScript your pages reference, measuring the files it can fetch, and comparing the results over time. Useful when you want to see what a plugin update or site change has added.
IndexLane Site Atlas
WordPress Version 0.1.0
A WordPress plugin for figuring out how a site fits together before you start changing things. It looks at the connections between plugins, content and stored data, which is useful when you’re planning a cleanup, rebuild or handover.
Food Cost & Profit Tracker
TypeScript Version 2.2.2
A private app for tracking food costs, stock and estimated profit across direct sales and food delivery. It connects ingredient purchases, recipes, orders and platform fees, helping a seller understand what each dish costs and how much money is actually left.
Comfy Money - Private Multi-Currency Ledger
Go Version 0.14.5
Comfy Money is a private money tracker I built for myself and my family. It handles wallets in different currencies, income, expenses, transfers, budgets and debts, with automatically updated exchange rates and a mobile interface that installs as a PWA.
crawlcontract
crawlcontract
Rust Version 0.3.1
A command-line tool for catching crawl and indexing problems when a website changes. It checks built files or pages fetched from a live or preview site, and compares scans so you can spot problems before a release.
Private setup for GitHub Actions runners
GitHub Actions Version 0.1.2
GHA is my private setup for running GitHub Actions runners for multiple repositories on one VDS. It handles runner registration, a shared job queue and resource limits, letting me run CI jobs on my own server while keeping the workflows in GitHub.
Bug Squasher: DOM Defense
TypeScript Version 0.1.0
I was bored and wanted a break from client projects, so I made a little TypeScript game. Naturally, it ended up being about bugs taking down a server.
DG6 — Multilingual pSEO Publishing
PHP Version 1.0.0
A Laravel publishing application for my programmatic SEO projects. DG6 turns generated content into multilingual websites, with separate domains, themes and publishing schedules, all running from one codebase.
DG6 Swarm — AI Content & Media Pipeline
Python Version 0.17
A Python pipeline that turns a domain and starting keyword into site copy, articles and visual assets. Specialized GPU workers handle each stage, starting and stopping according to queued work.