Ollama structured output bug skips schema when thinking models answer directly
Ollama versions since 0.34.4 fail to enforce JSON schemas on thinking models that skip the reasoning step, returning unstructured text with HTTP 200.
News for teams that run their own software
A daily digest on self-hosting, DevOps, security, document AI and HR tech. Each story is summarized, explains why it matters and links to the original publisher.
Grading AI agents on database state instead of text output
Microsoft and Hugging Face released ThinkingBox, a framework that evaluates AI agents by verifying their final database records rather than their conversational logs.
Aura brings scheduled AI agents and graph memory to self-hosted stacks
A new Go-based agent runtime offers temporal knowledge graphs, scheduled jobs, and multi-channel access via Telegram and WhatsApp for self-hosted environments.
Browser Use vs Skyvern: verifying self-hosted AI agent completion
A technical comparison of Browser Use and Skyvern reveals that schema validation alone cannot verify workflow success. Teams must check artifact integrity independently.
Running Coolify on a $6 VPS: Benchmarks and swap file requirements
A timed benchmark shows Coolify runs on a 1 GB DigitalOcean droplet if you add a 2 GB swap file, with install taking 172 seconds.
MentaAgent brings self-hosted AI business analysis to local Docker setups
Daniel Kim released MentaAgent, an open-source tool that runs locally to analyze business documents and answer questions with source citations.
Homelab census reveals routing logic for AI agents on limited hardware
A detailed inventory of a four-node homelab shows how 41 containers and a single 6 GB GPU dictate AI agent routing between local execution and cloud APIs.
Lightweight Rust GUI brings container management to Windows 11 WSL
RC Desktop offers a native, low-resource alternative to Electron-based tools for managing WSL containers on Windows 11.
Aleph Alpha releases Kolibri, a sovereign German-English LLM
Aleph Alpha launched Kolibri, an open-weight mixture-of-experts model trained in Europe. It prioritizes data sovereignty and efficient German language processing for self-hosted deployments.
Microsoft and Hugging Face benchmark AI agent reliability on database state
A new benchmark reveals that AI agents often report success while leaving incorrect database records, highlighting a gap between tool calls and actual outcomes.
Air-gapped AI architecture for industrial safety in Pakistan
An oil and gas operator in Pakistan deployed a fully self-hosted AI stack to predict safety incidents without sending data to the cloud.
Separating heartbeat monitoring from job logic for safer rollbacks
A technical guide explains why scheduled jobs need external heartbeat monitors to detect silent failures and ensure safe rollbacks during deployment.
Android CLI adds remote device streaming and specialized AI skills
Google updates Android CLI with remote physical device access and new AI skills for Wear OS, security, and compliance.
Cloudflare Traces brings end-to-end request visibility to open beta
Cloudflare launches Traces in open beta, offering automatic OpenTelemetry-based tracing for requests across its platform with configurable sampling and OTLP export.
Portus gateway swaps Pingora for Rama engine to cut memory use
Portus version 0.2.4 adopts the Rama network stack, delivering up to 31 percent higher throughput and significantly lower memory usage compared to its previous Pingora base.
