One-line CI fix resolves pgvector extension mismatch in FastAPI RAG backend

·1 views

A FastAPI-based AI support-agent project called mini-agent experienced a CI failure on August 12, 2026, when its GitHub Actions workflow could not complete the test suite. The failure occurred because the workflow used a generic postgres:17-alpine container image, which lacked the pgvector extension required by the application's database schema. The fix involved replacing that single image reference with pgvector/pgvector:0.8.6-pg17, leaving all other workflow configuration unchanged. The updated CI run completed successfully, with all tests passing where they had previously failed. The incident highlights the importance of treating database extensions as explicit runtime dependencies in integration test environments.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer releases Kintara, an open-source self-hosted document library with optional AI tools

A developer has launched Kintara, a self-hosted document management system that runs in Docker and automatically indexes PDFs, Markdown, and text files from a watched folder. The app offers full-text search, tags, collections, reading progress, highlights, and is accessible via browser or installable PWA across multiple devices. Built on a single Rust server handling both the API and frontend, Kintara evolved significantly during development, replacing an earlier Tauri desktop shell. Optional AI integrations via OpenAI or Gemini enable document summarization, metadata suggestions, cover image generation, and a citation-verified document search feature. The project is open source, with the code available on GitHub and a Docker image published on Docker Hub.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Free AI Token Quotas Can Cost Teams More Than Paid Plans

Developers using free AI token allowances often underestimate the true costs of model experiments, which include integration, evaluation, and cleanup work that never appears on an invoice. Common failure modes include unbounded retry loops that drain quotas rapidly, silent vendor lock-in from provider-specific code, and shared budgets consumed by a single runaway script. A proposed solution is a lightweight "gated client" — a small Python wrapper that enforces a hard token budget and timeout on any OpenAI-compatible endpoint. The script aborts and logs a message the moment spending limits are reached, making cost overruns visible rather than silent. The author argues that treating free allowances as finite, metered resources from the start prevents expensive rework during or after the experiment phase.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

LLM Model Fingerprinting: Why AI Teams Must Verify What Their Gateway Serves

As AI stacks grow more complex — with gateways, fallbacks, proxies, and tenant-specific routing — developers can no longer rely on a model's self-reported identity to confirm what is actually running in production. A technique called LLM model fingerprinting uses infrastructure-level signals such as token counts, context limits, tool-call formatting, and latency profiles to verify that the correct model is being served. Unlike prompt-based identification, which can be spoofed by fine-tunes or system instructions, these behavioral artifacts are significantly harder to fake. The approach functions as a smoke test for AI infrastructure, helping teams detect silent model swaps, unintended fallback activations, and configuration drift before they affect customer-facing workflows. Fingerprinting is not a replacement for evaluations but a complement — ensuring that benchmark results and production serving reflect the same underlying model.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Dev Guide: How to Build Local-Service Sites That Answer Customer Queries Fast

A technical guide published on DEV Community outlines how developers can build local-service websites — using an RV repair business as a case study — that quickly answer the four questions customers ask most: service coverage, location, availability, and next steps. The core argument is that losing a service call is a systems-design problem, not a marketing failure, stemming from inconsistent or inaccessible business data across a site. The guide recommends centralizing all business facts — phone numbers, service areas, hours, and repair categories — into a single canonical data model that feeds service pages, JSON-LD, sitemaps, and Google Business Profile simultaneously. It also introduces a code-level eligibility check to prevent the automated creation of thin, location-service landing pages that lack real supporting evidence, which Google flags as doorway page spam. The approach prioritizes server-rendered content so neither users nor search crawlers have to wait on client-side API calls to access critical business information.

0 comments Read more at DEV Community